bluetooth/wlan receiver design methodology … · bluetooth/wlan receiver design methodology and ic...
TRANSCRIPT
BLUETOOTH/WLAN RECEIVER DESIGN METHODOLOGY AND
IC IMPLEMENTATIONS
A Dissertation
by
AHMED AHMED ELADAWY EMIRA
Submitted to the Office of Graduate Studies of Texas A&M University
in partial fulfillment of the requirements for the degree of
DOCTOR OF PHILOSOPHY
December 2003
Major Subject: Electrical Engineering
BLUETOOTH/WLAN RECEIVER DESIGN METHODOLOGY AND
IC IMPLEMENTATIONS
A Dissertation
by
AHMED AHMED ELADAWY EMIRA
Submitted to Texas A&M University in partial fulfillment of the requirements
for the degree of
DOCTOR OF PHILOSOPHY
Approved as to style and content by:
Edgar Sánchez-Sinencio
(Chair of Committee)
José Silva-Martínez (Member)
Aydin Karsilayan
(Member)
Scott Miller (Member)
Duncan Walker
(Member)
Chanan Singh (Head of Department)
December 2003
Major Subject: Electrical Engineering
iii
ABSTRACT
Bluetooth/WLAN Receiver Design Methodology and
IC Implementations. (December 2003)
Ahmed Ahmed Eladawy Emira, B.Eng., Cairo University, Egypt;
M.Eng., Cairo University, Egypt
Chair of Advisory Committee: Dr. Edgar Sánchez-Sinencio
Emerging technologies such as Bluetooth and 802.11b (Wi-Fi) have fuelled the
growth of the short-range communication industry. Bluetooth, the leading WPAN
(wireless personal area network) technology, was designed primarily for cable
replacement applications. The first generation Bluetooth products are focused on
providing low-cost radio connections among personal electronic devices. In the WLAN
(wireless local area network) arena, Wi-Fi appears to be the superior product. Wi-Fi is
designed for high speed internet access, with higher radio power and longer distances.
Both technologies use the same 2.4GHz ISM band. The differences between Bluetooth
and Wi-Fi standard features lead to a natural partitioning of applications. Nowadays,
many electronics devices such as laptops and PDAs, support both Bluetooth and Wi-Fi
standards to cover a wider range of applications. The cost of supporting both standards,
however, is a major concern. Therefore, a dual-mode transceiver is essential to keep the
size and cost of such system transceivers at a minimum.
iv
A fully integrated low-IF Bluetooth receiver is designed and implemented in a low
cost, main stream 0.35µm CMOS technology. The system includes the RF front end,
frequency synthesizer and baseband blocks. It has -82dBm sensitivity and draws 65mA
current. This project involved six Ph.D. students and I was in charge of the design of the
channel selection complex filter.
In the Bluetooth transmitter, a frequency modulator with fine frequency steps is
needed to generate the GFSK signal that has ±160kHz frequency deviation. A low power
ROM-less direct digital frequency synthesizer (DDFS) is designed to implement the
frequency modulation. The DDFS can be used for any frequency or phase modulation
communication systems that require fast frequency switching with fine frequency steps.
Another contribution is the implementation of a dual-mode 802.11b/Bluetooth
receiver in IBM 0.25µm BiCMOS process. Direct-conversion architecture was used for
both standards to achieve maximum level of integration and block sharing. I was honored
to lead the efforts of seven Ph.D. students in this project. I was responsible for system
level design as well as the design of the variable gain amplifier. The receiver chip
consumes 45.6/41.3mA and the sensitivity is -86/-91dBm.
v
DEDICATION
To my Parents,
To my Brother, and Sister,
And to my fiancée Eman
For their love and support
vi
ACKNOWLEDGEMENTS
A journey is easier when you travel together. Interdependence is certainly more
valuable than independence. This dissertation is the result of four years of work whereby
I have been accompanied and supported by many people. It is a pleasant aspect to now
have the opportunity to express my gratitude for all of them.
With a deep sense of gratitude, I wish to express my sincere thanks to my supervisor,
Dr. Edgar Sánchez-Sinencio, for his immense help in planning and executing the work in
a timely manner. The confidence and dynamism with which Dr. Sánchez guided the work
requires no elaboration. His company and assurance at the time of crisis will always be
remembered. His technical and editorial advice was essential to the completion of this
dissertation and taught me innumerable lessons and insights on the workings of academic
research in general.
I am also grateful to Dr. Sherif Embabi and his group at Texas Instruments for
sharing much of their valuable time to review the design of the Bluetooth and Chameleon
receiver projects. Their comments and suggestions were very useful in avoiding potential
design problems and improving the performance of the receiver. I would also like to
express my gratitude to Dr. José Silva-Martínez and Dr. Aydin Karsilayan who were
always there to answer my technical questions throughout my entire Ph.D. program.
Thanks also go to my committee members: Dr. Scott Miller and Dr. Duncan Walker for
their time and valuable comments on my research.
I am fortunate to have had the opportunity to work with a group of energetic people
in the Analog and Mixed Signal Center (AMSC). Special thanks and words of
appreciation are due to my friend, research partner, and ex roommate Ahmed Nader
vii
Mohieldin. I was his roommate during the first two years of my Ph.D. He made my
arrival to College Station a lot smoother than I thought it would be. He was like my
family away from home. On the technical side, Ahmed and I worked together on three
fruitful projects, which constituted a major part of my Ph.D. I learned much from our
numerous technical discussions. I also enjoyed the company of many students with whom
I had very interesting technical discussions. I had the most wonderful learning experience
working with a team of very active and bright Ph.D. students in the Bluetooth and the
Chameleon projects, which included Alberto Valdes-Garcia, Ari Valero-Lopez, Wenjun
Sheng, Bo Xia, Sung Tung Moon, Chunyu Xin, and Ahmed Nader Mohieldin. Their
talent and determination was behind the success of the two projects. I am also grateful to
my friend and colleague, Faisal Abdellatif Elseddeek, for good times we spent together as
roommates. He provided me the emotional support I much needed at the last stage of my
Ph.D.
Finally, I will never find words enough to express the gratitude that I owe to my late
father, my mother, my sister Sahar, my brother Mohamed, and my fiancée Eman. Their
tender love and support have always been the cementing force for building the blocks of
my career. The all round support rendered by them provided the much needed stimulant
to sail through the phases of stress and strain.
viii
TABLE OF CONTENTS
Page
ABSTRACT ..................................................................................................................iii
DEDICATION................................................................................................................ v
ACKNOWLEDGEMENTS............................................................................................vi
TABLE OF CONTENTS .............................................................................................viii
LIST OF FIGURES........................................................................................................xi
LIST OF TABLES .....................................................................................................xviii
CHAPTER
I INTRODUCTION ............................................................................................1
1.1. Background and Motivation ....................................................................1 1.2. Dissertation Overview.............................................................................2
II SHORT-RANGE COMMUNICATION STANDARDS....................................4
2.1. Short-Range Wired Communications ......................................................5 2.2. Short-Range Wireless Communications...................................................5 2.3. What is Bluetooth? ..................................................................................8
2.3.1. Bluetooth Operation................................................................... 11 2.3.2. Modulation Format .................................................................... 13 2.3.3. Frequency Band......................................................................... 15 2.3.4. Blocking Requirements.............................................................. 17 2.3.5. Intermodulation Requirements ................................................... 17 2.3.6. Sensitivity.................................................................................. 18
2.4. What is 802.11b?................................................................................... 18 2.4.1. Wi-Fi Operation......................................................................... 20 2.4.2. Modulation Format .................................................................... 22 2.4.3. Operating Frequency Range....................................................... 24 2.4.4. Blocking Requirements.............................................................. 24 2.4.5. Intermodulation Requirements ................................................... 24 2.4.6. Sensitivity.................................................................................. 25
2.5. Comparing Bluetooth and Wi-Fi............................................................ 25 2.6. What is UWB? ...................................................................................... 27
ix
CHAPTER Page
III BLUETOOTH RECEIVER ARCHITECTURE AND CHANNEL SELECTION FILTER .................................................................................... 31
3.1. Direct-Conversion Receiver Architecture .............................................. 32 3.1.1. DC Offsets................................................................................. 32 3.1.2. 1/f Noise.................................................................................... 33 3.1.3. Even Order Distortion................................................................ 34 3.1.4. I/Q Mismatch............................................................................. 35
3.2. Low-IF Receiver Architecture ............................................................... 36 3.3. Bluetooth Receiver Architecture............................................................ 39 3.4. Channel Select Complex Filter Design .................................................. 42
3.4.1. Complex Filter Theory............................................................... 42 3.4.2. Complex Filter Implementation ................................................. 46 3.4.3. Frequency Tuning Scheme......................................................... 60 3.4.4. Experimental Results ................................................................. 63
3.5. Conclusions........................................................................................... 68
IV DIRECT DIGITAL FREQUENCY SYNTHESIZER (DDFS)......................... 69
4.1. Conventional DDFS Architecture.......................................................... 70 4.2. Proposed DDFS Architecture ................................................................ 74 4.3. Linear DAC........................................................................................... 79 4.4. Switched Weighted-Sum Block ............................................................. 81 4.5. Experimental Results............................................................................. 88 4.6. Particular Case ...................................................................................... 96 4.7. Conclusions........................................................................................... 97
V CHAMELEON: A MULTI-STANDARD RECEIVER DESIGN.................... 99
5.1. Possible Receiver Architectures............................................................. 99 5.2. Proposed Direct-Conversion Receiver Architecture............................. 102 5.3. System Design Issues .......................................................................... 104
5.3.1. BER versus SNR ..................................................................... 105 5.3.2. DC Offset Problem .................................................................. 108 5.3.3. Filter Frequency Response....................................................... 117 5.3.4. ADC Resolution and Sampling Rate ........................................ 120 5.3.5. VGA Dynamic Range .............................................................. 123 5.3.6. I/Q Mismatches ....................................................................... 124 5.3.7. Even-order Distortion .............................................................. 127 5.3.8. VCO Frequency Pulling........................................................... 127 5.3.9. Combined Effects .................................................................... 128
5.4. From Standard to Block Specifications................................................ 130 5.4.1. From Standard to Receiver Specifications................................ 131 5.4.2. Determination of Receiver Building Blocks Specifications ...... 138 5.4.3. System Design Verification ..................................................... 148
x
CHAPTER Page
VI CHAMELEON: CIRCUIT IMPLEMENTATIONS ...................................... 158
6.1. LNA and Mixer ................................................................................... 158 6.2. VCO and PLL ..................................................................................... 161 6.3. Channel Select Filter ........................................................................... 166 6.4. Variable Gain Amplifier ...................................................................... 171
6.4.1. DC Offset Problem, Revisited.................................................. 171 6.4.2. Proposed Solution for HPF Slow Response.............................. 173 6.4.3. OpAmp and Buffer Designs..................................................... 175 6.4.4. VGA Gain Lineup ................................................................... 178 6.4.5. Testing Results ........................................................................ 182
6.5. Analog to Digital Converter ................................................................ 186
VII CHAMELEON: EXPERIMENTAL RESULTS ........................................... 190
7.1. System Chip ........................................................................................ 190 7.2. Experimental Results........................................................................... 192
7.2.1. BER vs. Input Power (Sensitivity) ........................................... 192 7.2.2. IIP3 ......................................................................................... 195 7.2.3. IIP2 ......................................................................................... 197
VIII CONCLUSIONS ........................................................................................ 199
REFERENCES............................................................................................................ 201
APPENDIX A............................................................................................................. 207
APPENDIX B ............................................................................................................. 209
APPENDIX C ............................................................................................................. 214
APPENDIX D............................................................................................................. 217
APPENDIX E ............................................................................................................. 232
VITA........................................................................................................................... 235
xi
LIST OF FIGURES
Page
Fig. 2.1 Short range communication standards ...............................................................4
Fig. 2.2 Point-to-point and point-to-multipoint and scatternet topology in Bluetooth.... 13
Fig. 2.3 Gaussian LPF impulse response ...................................................................... 14
Fig. 2.4 Gaussian LPF frequency response................................................................... 14
Fig. 2.5 GFSK modulation steps .................................................................................. 16
Fig. 2.6 Frequency hopping diagram............................................................................ 17
Fig. 2.7 An ad hoc wireless network with three stations using Wi-Fi............................ 21
Fig. 2.8 A simple infrastructure network using Wi-Fi................................................... 21
Fig. 2.9 Construction of the CCK modulated signal ..................................................... 22
Fig. 2.10 Comparison of spatial density of Bluetooth, Wi-Fi, and UWB ........................ 30
Fig. 3.1 Direct-conversion receiver architecture. .......................................................... 32
Fig. 3.2 Effect of even order distortion......................................................................... 35
Fig. 3.3 IF receiver architecture. .................................................................................. 37
Fig. 3.4 Down-conversion with a single sinusoidal signal. ........................................... 37
Fig. 3.5 Down-conversion with a single exponential. ................................................... 38
Fig. 3.6 Basic low-IF receiver architecture................................................................... 39
Fig. 3.7 A Low-IF Bluetooth receiver architecture [15]................................................ 41
Fig. 3.8 Receiver image rejection architecture in the complex domain. ........................ 43
Fig. 3.9 Frequency translation of a complex (quadrature) mixer (a) before complex mixing (signal A in Fig. 3.8) (b) after complex mixing (signal B in Fig. 3.8)... 45
Fig. 3.10 Practical implementation of the receiver image rejection ................................ 46
Fig. 3.11 LPF shifted to ωIF (a) conceptual complex representation (b) actual building block implementation (c) Active-RC implementation...................................... 48
xii
Page
Fig. 3.12 Pole locus of the LPF prototype and the complex BPF .................................... 49
Fig. 3.13 Fully differential active-RC complex first-order LPF ...................................... 49
Fig. 3.14 Linear frequency translation to convert LPF to complex BPF.......................... 50
Fig. 3.15 Differential OTA-C implementation of linear frequency translation ................ 50
Fig. 3.16 Pseudo differential OTA ................................................................................. 51
Fig. 3.17 (a) No CM control (b) CMFB circuit (c) CMFF circuit .................................. 52
Fig. 3.18 (a) Biasing circuit (b) The CMD and auxiliary circuit required for CMFF and CMFB...................................................................................................... 53
Fig. 3.19 (a) I branch of the complex biquadratic section (b) Conceptual complex biquad............................................................................................................. 54
Fig. 3.20 (a) Circuit setup for HD3 analysis (b) Another scenario for HD3 analysis ....... 57
Fig. 3.21 The complete 12th order complex filter............................................................ 59
Fig. 3.22 Frequency tuning circuit for complex filter ..................................................... 61
Fig. 3.23 The relaxation oscillator.................................................................................. 63
Fig. 3.24 The die photo (filter area=1.6×0.8mm2 and tuning circuit area=1×0.4mm2)..... 65
Fig. 3.25 Test setup for the complex filter...................................................................... 65
Fig. 3.26 Frequency response at signal and image sides (vertical axis 12dB/div, … ideal, actual).............................................................................................. 66
Fig. 3.27 The measured SFDR versus n ......................................................................... 66
Fig. 3.28 IM3 test for f1 = 1.95MHz and f2 = 2.05MHz.................................................. 67
Fig. 3.29 Group delay response...................................................................................... 67
Fig. 4.1 Bluetooth transmitter architecture ................................................................... 69
Fig. 4.2 Conventional ROM-based DDFS .................................................................... 71
Fig. 4.3 Compression technique using approximation f(x) of the sine function............. 72
Fig. 4.4 Combining a small ROM and linear interpolation to generate sine function..... 73
xiii
Page
Fig. 4.5 ROM-Less DDFS using non-linear DAC ........................................................ 73
Fig. 4.6 Sine wave approximation................................................................................ 74
Fig. 4.7 Block diagram of the proposed DDFS architecture.......................................... 76
Fig. 4.8 Effect of the number of segments on the SFDR............................................... 77
Fig. 4.9 Effect of phase resolution on the SFDR........................................................... 77
Fig. 4.10 Effect of finite unit resistance on the SFDR .................................................... 78
Fig. 4.11 Linear DAC circuit implementation ................................................................ 80
Fig. 4.12 Switched weighted-sum block......................................................................... 81
Fig. 4.13 Implementation of the weighted-sum block..................................................... 83
Fig. 4.14 Alternative implementation of the weighted-sum block................................... 86
Fig. 4.15 Chip micrograph ............................................................................................. 90
Fig. 4.16 Testing setup................................................................................................... 90
Fig. 4.17 Single-ended and differential outputs of the I branch (fout=98kHz) .................. 91
Fig. 4.18 Quadrature outputs I and Q at fout=98kHz ....................................................... 91
Fig. 4.19 Frequency modulation..................................................................................... 92
Fig. 4.20 Amplitude modulation .................................................................................... 92
Fig. 4.21 Output spectrum at fclk=100MHz and fout=98kHz with SFDR=57.3dBc........... 93
Fig. 4.22 Output spectrum at fclk=100MHz and fout=1.56MHz with SFDR=42.1dBc ...... 93
Fig. 4.23 SFDR versus synthesized frequency at fclk=100MHz....................................... 94
Fig. 4.24 SFDR versus clock frequency at fout= fclk/128.................................................. 94
Fig. 4.25 Front end of DDFS ......................................................................................... 97
Fig. 5.1 Dual mode receiver architecture used in [37]................................................. 101
Fig. 5.2 Dual mode receiver architecture used in [38]................................................. 102
Fig. 5.3 Dual-mode 802.11b/Bluetooth receiver. ........................................................ 103
xiv
Page
Fig. 5.4 Basic Wi-Fi simulation setup in System View ................................................ 106
Fig. 5.5 Bluetooth GFSK non-coherent demodulator used in MATLAB .................... 106
Fig. 5.6 BER versus SNR in Bluetooth mode in the ideal case (Sampling rate = 20MSample/s)............................................................................................... 107
Fig. 5.7 BER versus SNR performance of Wi-Fi 11Mbit/s CCK demodulator (Sampling rate = 88 MSample/s)................................................................... 107
Fig. 5.8 Self-mixing of (a) LO. (b) Interferers. ........................................................... 108
Fig. 5.9 BER performance of Bluetooth GFSK demodulator for different DC offsets (percentage relative to the peak signal value) ................................................ 111
Fig. 5.10 BER performance of Wi-Fi 11Mbit/s CCK demodulator for different DC offsets percentage (relative to the peak signal value) ..................................... 111
Fig. 5.11 Power spectral density of Bluetooth GFSK signal ......................................... 113
Fig. 5.12 Power spectral density of Wi-Fi CCK signal ................................................. 113
Fig. 5.13 Bluetooth GFSK demodulator performance with two HPFs .......................... 114
Fig. 5.14 Wi-Fi 11Mb/s CCK demodulator performance with four HPFs ..................... 114
Fig. 5.15 Analog offset cancellation loop using feedback integrator [41, 42]................ 116
Fig. 5.16 Offset cancellation loop using DAC [43]....................................................... 117
Fig. 5.17 BER performance using different LPF approximations in Bluetooth mode.... 118
Fig. 5.18 Bluetooth BER performance using Butterworth LPF with different bandwidths ................................................................................................... 119
Fig. 5.19 Wi-Fi 11Mbit/s CCK BER using different LPF approximations .................... 119
Fig. 5.20 Wi-Fi BER performance using Butterworth LPF with different bandwidths .. 120
Fig. 5.21 Bluetooth GFSK BER performance for different ADC quantization bits ....... 121
Fig. 5.22 Bluetooth GFSK BER performance for different ADC sampling rates .......... 122
Fig. 5.23 ADC quantization effect on BER performance of Wi-Fi 11Mb/s CCK.......... 122
Fig. 5.24 Wi-Fi 11Mb/s CCK BER performance for different ADC sampling rates...... 123
xv
Page
Fig. 5.25 BER performance of Bluetooth demodulator for different gain mismatches .. 125
Fig. 5.26 BER performance of Bluetooth demodulator for different phase mismatches 125
Fig. 5.27 Wi-Fi 11Mb/s CCK BER performance for different I/Q gain mismatches. .... 126
Fig. 5.28 Wi-Fi 11Mb/s CCK BER performance for different I/Q phase mismatches. .. 126
Fig. 5.29 Injection pulling due to large interferer ......................................................... 128
Fig. 5.30 Effects of combined receiver non-idealities on Bluetooth BER performance . 129
Fig. 5.31 Effects of combined receiver non-idealities on Wi-Fi BER performance ....... 130
Fig. 5.32 Flowchart of the system design process......................................................... 131
Fig. 5.33 Noise contributions from the different receiver blocks in Bluetooth mode..... 142
Fig. 5.34 Noise contributions from the different receiver blocks in Wi-Fi mode........... 142
Fig. 5.35 IIP3 contributions from receiver blocks in both modes.................................. 143
Fig. 5.36 IIP2 contributions from receiver blocks in both modes.................................. 144
Fig. 5.37 LO phase noise effect.................................................................................... 145
Fig. 5.38 ADC input level versus RF input power in Bluetooth mode .......................... 149
Fig. 5.39 Gain switching strategy in Bluetooth mode ................................................... 149
Fig. 5.40 Signal and noise levels in Bluetooth sensitivity test....................................... 150
Fig. 5.41 Signal-to-noise ratio in Bluetooth sensitivity test........................................... 150
Fig. 5.42 Adjacent channel test in Bluetooth mode....................................................... 151
Fig. 5.43 Bluetooth maximum signal test ..................................................................... 151
Fig. 5.44 Signal test at gain switching point (LNA gain is activated and VGA gain is low) .............................................................................................................. 152
Fig. 5.45 Signal test at gain switching point (LNA is bypassed and VGA gain is low) . 152
Fig. 5.46 ADC input level versus RF input power in Wi-Fi mode ................................ 154
Fig. 5.47 Gain switching strategy in Wi-Fi mode ......................................................... 154
Fig. 5.48 Signal and noise levels in Wi-Fi sensitivity test............................................. 155
xvi
Page
Fig. 5.49 Signal-to-noise ratio in Wi-Fi sensitivity test................................................. 155
Fig. 5.50 Adjacent channel test in Wi-Fi mode............................................................. 156
Fig. 5.51 Wi-Fi maximum signal test ........................................................................... 156
Fig. 5.52 Signal test at gain switching point in Wi-Fi mode (LNA gain is activated) .... 157
Fig. 5.53 Signal test at gain switching point in Wi-Fi mode (LNA is bypassed) ........... 157
Fig. 6.1 LNA circuit................................................................................................... 159
Fig. 6.2 Mixer circuit ................................................................................................. 160
Fig. 6.3 Input matching for the high gain mode.......................................................... 160
Fig. 6.4 Input matching for the low gain mode ........................................................... 161
Fig. 6.5 Cascode charge pump schematic. .................................................................. 162
Fig. 6.6 Capacitance multiplier schematic. ................................................................. 162
Fig. 6.7 VCO schematic............................................................................................. 164
Fig. 6.8 Phase noise of the LO ................................................................................... 164
Fig. 6.9 Phase switching prescaler.............................................................................. 166
Fig. 6.10 Butterworth filter block diagram ................................................................... 168
Fig. 6.11 Biquadratic section ....................................................................................... 169
Fig. 6.12 Implementation of the dual-mode OTA......................................................... 169
Fig. 6.13 Dual-mode operation and programmability of the LPF.................................. 170
Fig. 6.14 DC offset correction using digital feedback................................................... 173
Fig. 6.15 DC offset correction using passive HPF ........................................................ 173
Fig. 6.16 (a) Conventional OpAmp-R VGA (b) Proposed constant output offset VGA 175
Fig. 6.17 Proposed constant output offset VGA circuit using OpAmp-R ...................... 175
Fig. 6.18 3-stage OpAmp schematic ............................................................................ 177
Fig. 6.19 CMFB circuit ................................................................................................ 177
xvii
Page
Fig. 6.20 The buffer circuit .......................................................................................... 178
Fig. 6.21 Second stage VGA circuit ............................................................................. 180
Fig. 6.22 Third stage VGA circuit................................................................................ 181
Fig. 6.23 VGA test setup.............................................................................................. 184
Fig. 6.24 Time domain response of the VGA ............................................................... 184
Fig. 6.25 Settling time of the VGA .............................................................................. 185
Fig. 6.26 Frequency response of the VGA at different gain settings ............................. 185
Fig. 6.27 Time-interleaved pipeline ADC architecture ................................................. 188
Fig. 6.28 Online digital calibration............................................................................... 189
Fig. 6.29 Measured SNR of the ADC in (a) Bluetooth mode (64dB at 11Msample/s) (b) WiFi mode (48dB at 44Msample/s) ......................................................... 189
Fig. 7.1 Chameleon die micrograph ........................................................................... 191
Fig. 7.2 Simplified block diagram of the testing board for the Chameleon chip .......... 191
Fig. 7.3 Photograph of the PCB used to test the Chameleon chip ............................... 192
Fig. 7.4 Testing setup for NF measurement ................................................................ 194
Fig. 7.5 BER performance of the dual-mode receiver................................................. 194
Fig. 7.6 Testing setup for IIP3/IIP2 two-tone test ....................................................... 196
Fig. 7.7 Receiver IIP3 when LNA is in high gain mode (VGA gain = 12dB).............. 196
Fig. 7.8 Receiver IIP2 when LNA is in high gain mode (VGA gain = 24dB).............. 197
xviii
LIST OF TABLES
Page
Table 2.1 Summary of characteristics of some leading WLAN/WPAN standards ..........7
Table 2.2 Bluetooth radio specifications ...................................................................... 12
Table 2.3 802.11b radio specifications ......................................................................... 20
Table 2.4 DQPSK encoding table ................................................................................ 23
Table 2.5 5.5Mb/s CCK encoding table ....................................................................... 23
Table 2.6 QPSK encoding table .................................................................................... 24
Table 3.1 Summarized filter testing results .................................................................. 68
Table 4.1 Required values of (M), (∆R/Rmax), and corresponding SFDR for different number of segments ..................................................................................... 79
Table 4.2 Comparison with recently published work.................................................... 95
Table 5.1 Receiver parameters and non-idealities used in the BER simulations.......... 129
Table 5.2 Summary of receiver specifications............................................................ 138
Table 5.3 Receiver gain distribution .......................................................................... 140
Table 5.4 NF, IIP3, and IIP2 distribution in Bluetooth/Wi-Fi mode ........................... 141
Table 5.5 Required LO phase noise in Bluetooth mode.............................................. 146
Table 6.1 Summary of performance........................................................................... 170
Table 6.2 Summary of required VGA specifications .................................................. 171
Table 6.3 Gain values of the second stage and the corresponding digital inputs ......... 180
Table 6.4 Gain values of the third stage and the corresponding digital inputs............. 182
Table 6.5 Gain distribution over the three VGA stages .............................................. 183
Table 6.6 Summary of VGA experimental results ...................................................... 186
Table 7.1 Receiver area breakdown ........................................................................... 190
Table 7.2 Performance summary of the Chameleon receiver...................................... 198
xix
Page
Table 7.3 Power consumption contribution of the receiver blocks.............................. 198
1
CHAPTER I
INTRODUCTION
Stimulated by the increased consumer and commercial users demand or wireless
communications applications, the advancement of sophistication in wireless circuit
design has progressed at an unprecedented speed in recent years. The last few years have
witnessed a remarkable miniaturization of portable equipment, a similar extension of
battery life, and almost an order of magnitude in retail prices. Two ubiquitous examples
of this phenomenon are the wireless local area networks (WLANs) and wireless personal
area networks (WPANs).
1.1. Background and Motivation
Although not well known among everyday wireless communication consumers,
commercial wireless communications technology has also experienced a similar impetus
in its advancement, both in miniaturization and cost. In portable wireless communication
industry it is well known that the design of the RF transceiver is usually the key element
that determines the cost, size, and useful battery life of the equipment, as well as how the
equipment is used.
One of the major forces sustaining continual research and development in the
wireless communication arena is the general user acceptance of wireless communications
standards. These standards provide coalescence within the industry to invest in the
creation of suitable chip sets for the manufacturing of consumer subscriber units. The
resulting high volume drives down the cost of the final product, and industry is able to
This dissertation follows the style of IEEE Journal of Solid-State Circuits.
2
take advantage of the same trends that have dramatically lowered prices in the PC
industry.
The main goal of this dissertation is to present system level and building block level
solutions for two of the most widespread short-range communications standards,
Bluetooth and 802.11b (Wi-Fi). The motivation is to bridge the gap between the circuit
and wireless vision in wireless design. Although the dissertation focuses on Bluetooth
and Wi-Fi radio implementations, many design aspects can be generalized to most other
radio implementations. The Bluetooth and Wi-Fi standards operate at RF frequency
2.4GHz that is also used by microwave ovens and cordless phones. While Wi-Fi is meant
to be wireless version of Ethernet, Bluetooth is designed to replace cables that are used
for relatively low speed data transfer. Wi-Fi offers data rates up to 11Mbit/s while
Bluetooth’s raw data rate is 1Mbits/s. Bluetooth is supposed to be lower cost than Wi-Fi.
In this dissertation, the design of a low-IF Bluetooth receiver fabricated in a low cost
mainstream CMOS technology is presented. On the other hand, there are many devices
that require having both standards implanted. In such cases, the integration of both
standards into a single chip would be cheaper than having two separate chip solutions.
For this purpose, a dual-mode Bluetooth/Wi-Fi chip solutions is also described.
1.2. Dissertation Overview
Chapter II presents short-range communications methods. Wired and wireless
communications standards are introduced with some focus on wireless standards since it
is the topic of the following chapters. In particular, details about Bluetooth and 802.11b
(Wi-Fi) standards are given to get the reader acquainted with these standards before
presenting the receivers design specifications and implementations in later chapters.
3
Chapter III presents Bluetooth possible receiver architectures and discusses the pros and
cons of each architecture. The proposed low-IF Bluetooth architecture is discussed with
some detail. As part of the Bluetooth receiver, detailed design, implementation, and
measurement results of channel selection complex filter are presented. In the Bluetooth
transmitter, on the other hand, direct digital frequency synthesizer (DDFS) is used as the
frequency modulator. DDFS design and implementation is discussed in details in chapter
IV. A dual-mode 802.11b/Bluetooth receiver design is presented in chapter V. The dual-
mode receiver is named “Chameleon” for its ability to change operation modes. An
overview of previous dual-mode receiver implementations is included. The proposed
direct conversion architecture is presented. Effects of receiver non-idealities of its
performance are simulated and quantified. Detailed design procedure starting from
standard specifications to receiver specifications to building block requirements is
presented. Chapter VI gives the design of the Chameleon receiver building blocks.
Variable gain amplifier (VGA) design is presented in details. Measured characteristics of
the receiver blocks are also included. In chapter VII, experimental results of the
Chameleon receiver are presented. Test setup as well as measurement procedures are
illustrated. Finally, chapter VIII summarizes the main contributions of this research work.
It should be mentioned that the Bluetooth and Chameleon projects were carried-out by a
team of seven Ph.D. students. This dissertation covers partially the entire design process
and implementations, but the details of some system aspects and building blocks
developed and proposed by the author are emphasized in the essence of this dissertation.
4
CHAPTER II
SHORT-RANGE COMMUNICATION STANDARDS
In recent years mobile digital devices such as personal digital assistants (PDAs),
mobile phones, digital cameras, and laptops have penetrated the consumer market. All
these devices require powerful short-range communication method for data exchange
among each other, connections with printers or local area network (LAN) access.
Basically, the communication methods can be based on cable connections, radio links, or
infrared links as illustrated in Fig. 2.1. Since each has its individual strengths and
weaknesses, each found its way into various products.
Short Range Communications Standards
WiredCommunications
WirelessCommunications
USB
Local areanetwork
Personal areanetwork
Bluetooth Home RF 802.11aUWB 802.11b 802.11g Fig. 2.1 Short range communication standards
5
2.1. Short-Range Wired Communications
Data exchange via cables is a well established method; universal serial bus (USB)
has become widely used standard interface. USB excels due to its high baud rates up to
480Mb/s, but suffers from its limited mobility due to cable connection. Therefore, USB is
best for applications that require stable high-performance connections for transmission of
high data volumes, where mobility is not very important. An example of application
would be the connection of a videoconferencing camera to your laptop.
2.2. Short-Range Wireless Communications
Contrary to USB, infrared transmission based on the Infrared Data Association
(IrDA) standard enables fast connection establishment due to its point-and-shoot
characteristic. Together with the high baud rates up to 16Mb/s, this makes transmission
well suited to applications that require high performance as hoc point-to-point
connections. Examples would include downloading of pictures from your digital camera
to your laptop or paying for your meal in your company’s cafeteria with your mobile
phone via IrDA port. IrDA standards have been there for decades and are widely
implemented in laptops, computers and PDAs. But until recently, either the cost was too
high, or in the case of infrared, the technology was too difficult to use.
Radio-based short-range wireless (SRW) communication is an alternative class of
emerging technologies designed primarily for indoor use over very short distances. It is
intended to provide fast (tens or hundreds of megabits per second) and low cost, cable-
free connections to the internet. SRW features transmission powers of several microwatts
6
up to milliwatts yielding a communication range between 10 and 100 meter. SRW will
provide connectivity to portable devices such as laptops, PDAs, cell phones and others.
Short-range communications standards fall into two broad but overlapping
categories: personal area networks (PAN) and local area networks (LAN).
Wireless PAN technologies emphasize low cost and low power consumption, usually
at the expense of range and peak speed. In a typical wireless PAN application, a short
wireless link, typically under 10 meters, replaces a computer serial cable or USB cable.
Standards, such as Bluetooth and HomeRF, have been created to regulate short-range
wireless communications. Bluetooth has appeared recently in many mobile devices.
Bluetooth can transmit data through solid nonmetal objects and supports a nominal link
range of 10cm-10m at a moderate baud rate up to 720kb/s (raw data rate is 1Mb/s) [1].
An optional high power mode in the current specifications allows for ranges up to 100m.
Because of the nature of radio, Bluetooth is a point to multipoint communication system,
which supports connections of two devices as well as ad hoc networking between several
devices. But in order to prevent unauthorized access, Bluetooth requires sophisticated
authentication and encryption mechanisms, which hamper fast connection establishment.
Therefore, Bluetooth is best for applications that require stable point-to-point or point-to-
multipoint connections for data exchange at moderate speeds, where mobility is a key
requirement. Ultra-wideband (UWB) is an emerging new technology that shows great
potential for SRW applications. Unlike conventional wireless communications systems
that are carrier-based, UWB-based communication is baseband. It uses a series of short
pulses that spread the energy of the signal from near DC to a few GHz.
7
Wireless LAN technologies, on the other hand, emphasize a higher peak speed and
longer range at the expense of cost and power consumption. Typically, wireless LANs
provide wireless links from portable laptops to a wired LAN access point. To date,
802.11b has gained acceptance rapidly as a wireless LAN standard. It has a nominal
open-space range of 100m and a peak over-the-air speed of 11Mb/s. Users can expect
maximum available speeds of about 5.5Mb/s. Other communication standards offer even
higher data rates, like 802.11a and 802.11g. Table 2.1 compares between the leading
radio-based short-range communication standards.
Table 2.1 Summary of characteristics of some leading WLAN/WPAN standards
Characteristic Bluetooth IEEE 802.11b IEEE 802.11g IEEE 802.11a UWB
Standard version/status
V 1.1 (Low-Rate) IEEE approved Draft IEEE
approved Draft
Maximum distance 10-100m 100m 100m 50m 10m
Frequency allocation
2.4GHz (ISM) 2.4GHz (ISM) 2.4GHz (ISM) 5GHz (UNII) 3.1-10.6GHz
Number of RF channels 79 3 3
12 (U.S.) 8 (EU)
4 (Japan) 1-15
Modulation GFSK QPSK (CCK) OFDM OFDM BPSK, QPSK
Spreading FH DSSS CCK OFDM OFDM (Multiband)
Maximum RF power 0-20dBm
30dBm (U.S.) 20dBm (EU)
10dBm (Japan)
30dBm (U.S.) 20dBm (EU)
10dBm (Japan)
17dBm, 24dBm, 30dBm
-41.3dBm/MHz
receiver sensitivity -70dBm -76dBm
for 11Mb/s -74dBm for
33Mb/s -65dBm or
54Mb/s -
8
2.3. What is Bluetooth?
Bluetooth technology was developed to create a short-range wireless voice and data
link between a broad range of devices such as PCs, notebook computers, handhelds and
PDAs (hereafter referred to as PDAs), Smart Phones, mobile phones and digital cameras.
Consistent with its aim of operating in even the smallest battery-powered devices, the
Bluetooth specification calls for a small form factor, low power consumption and low
cost. The range and speed of the technology were kept intentionally low so as to ensure
maximum battery life and minimum incremental cost for devices incorporating the
technology. At its heart, Bluetooth is about creating a Wireless Personal Area Network
(WPAN) consisting of all the Bluetooth-enabled electronic devices immediately
surrounding a user, wherever that user may be located. This project is supported by a
special interest group formed by hundreds of companies that lead technological
development.
Specifications of Bluetooth provide the limitations and possibilities of the
technology. Bluetooth operates in the 2.4GHz frequencies which is free for use to
everyone globally. This frequency provides an effective data rate of 720 Kbit/s. There are
multiple classes of transmission strengths which determine the use and the energy
efficiency of these devices. The Bluetooth network protocol employs five layers which
are implemented by software and hardware. They are:
(a) Application Programming Interface (API) – Allows the operating system (OS)
and other applications to access the Bluetooth interface.
(b) Logical Link Control and Adaptation Protocol (L2CAP) – Maintains individual
links to other devices in its transmission area.
9
(c) Link Manager – Responsible for establishing and terminating links and link
security.
(d) Baseband – low level tasks including error correction.
(e) Physical – actual radio transmitter and antenna.
The topology for Bluetooth is very unique and this feature allows it to be completely
scalable. The network link can be either point to point or can be point to multipoint, and
devices participating in the network can be defined as master or slave. In the simplest
network of one master device communicating with one slave device, a piconet is formed
and all the bandwidth is dedicated to the link between the two devices. Larger networks
can be formed with a single master and up to 7 slaves. The network architecture,
however, allows up to 255 slaves to be in a standby mode. Bluetooth manages to achieve
its functionality through its modulation and packeting scheme. This is also known as
Gaussian Frequency Shift Keying (GFSK).
Bluetooth hops to one of 79 different channels (US and Europe), a repetitive process
that keeps errors to a minimum. It is this packet hopping technique which leads to the
interference with 802.11b technology. The channel hopping technology of Bluetooth is
the secret behind its low power consumption, error correction and the distinct topology
that it can support. Bluetooth divides the data to be sent into packets. Each packet is sent
within a 625- microsecond slot. A frame is normally defined as a transmit and a receive
slot, providing full duplex communication between a master and a slave in one time
frame. To avoid noise and other interferences, Bluetooth hops to one of 79 different
channels each time frame. The channel that it hops to is determined by the Master ID and
the previous channel number. This algorithm is repeated again and again. For example if
10
there is severe noise between 2.408GHz and 2.410GHz, it will be avoided the majority of
the time. There is little to no time contention between masters within reach of each other
which might result in the masters picking up the same channels at the same time. Because
of this channel hopping mechanism, interference is kept to a minimum despite extremely
dense scatternets. Master devices can use this frame division to communicate with each
slave on the piconet consecutively, with 1 frame each, or they can devote multiple frames
to the same slave device, depending on the priority of the job at hand.
Bluetooth was originally conceived by Ericsson in 1994, when they began a study to
examine alternatives to cables that linked mobile phone accessories. Ericsson already
had a strong capability in short range wireless, having been a key pioneer of the European
DECT cordless telecommunications standard, which had been largely based upon their
earlier proprietary DCT900 technology. Out of their study was born the specification for
Bluetooth wireless.
Bluetooth was named after Harald Blatand (or Bluetooth), a tenth century Danish
Viking king who had united and controlled large parts of Scandinavia which are today
Denmark and Norway. The name was chosen to highlight the potential of the technology
to unify the telecommunications and computing industries - although it was chosen as an
internal codename, and it was never at the time expected to survive as the name used in
the commercial arena.
In February 1998, the Bluetooth SIG (Special Interest Group) was founded by a
small core of major companies - IBM, Intel, Nokia, Toshiba and Ericsson - to work
together to develop the technology and to subsequently promote its widespread
commercial acceptance.
11
Six months later the core Promoter Members publicly announced the global SIG and
invited other companies to join, with free access to the technology as Bluetooth adopters
in return for commitment to support the Bluetooth specification. Adoption was rapid and
1998-1999 saw a boom in the market for Bluetooth conference organizers, and vast
amounts of hype regarding the potential of the technology. In December 1999 it was
announced that four more major companies had joined the SIG as Promoter Members,
viz. Microsoft, Agere Systems (then Lucent), 3Com and Motorola.
The detailed Bluetooth specifications are available in [1]. Table 2.2 is a summary of
the key radio specifications.
2.3.1. Bluetooth Operation
Bluetooth controls timing on the network by designating one of the devices as a
master and the other as a slave. The master is simply the unit that initiates the
communication link, and the other participants are slaves. When that link is later broken,
the master/slave designations no longer apply. In fact, every Bluetooth device has both
master and slave hardware. The network itself is termed a piconet, meaning small
network. When there is only one slave, then the link is called point-to-point. A master can
control up to seven active slaves in a point-to-multipoint configuration. Slaves
communicate only with the master, never with each other directly. Timing is such that
members of the piconet cannot transmit simultaneously, so these devices will not jam
each other. Finally, communication across piconets can be realized if a Bluetooth device
can be a slave in more than one piconet, or a master in one and a slave in another.
Piconets configured in this manner are called scatternets. These various arrangements are
depicted in Fig. 2.2 [2].
12
Table 2.2 Bluetooth radio specifications
Frequency Band 2400 – 2483.5 MHz
Duplex Time Division
Modulation GFSK (BT = 0.5, Index: 0.28 – 0.35)
Channel Space 1 MHz
Sensitivity -70 dBm (for 0.1% BER)
Maximum Signal Level -20 dBm
C/Ico-channel 11 dB
C/I1MHz 0 dB
C/I2MHz -30 dB
C/I>=3MHz -40 dB
Interference
Performance C/Iimage -9 dB
30 MHz – 2000 MHz -10 dBm
2000 – 2399 MHz -27 dBm
2498 – 3000 MHz -27 dBm
Out-of-band Blocking
3000 MHz – 12.75 GHz -10 dBm
Interference Frequency 3, 4, 5 MHz
Interference Level -39 dBm
Intermodulation
Characteristics Bluetooth Signal Level 6 dB above sensitivity
Range -60 dBm±4 + 20±6 dB RSSI
Accuracy ±4 dB
Transmitted Frequency Accuracy ±75 kHz Radio Frequency
Tolerance Frequency Drift ±25 kHz in one slot
13
Master
SlaveSlave
Slave
Master
Slave
Slave Master
Point-to-multipointscatternet
Point-to-pointpiconet
scatternet member
Fig. 2.2 Point-to-point and point-to-multipoint and scatternet topology in Bluetooth
2.3.2. Modulation Format
The modulation format is Gaussian Frequency Shift Keying (GFSK) with a
bandwidth×bit time product (BT) of 0.5. The modulation index is between 0.28 and 0.35.
For the 1Mbps data rate in Bluetooth, frequency deviation can be from ±140 to ±175kHz.
The GFSK signal is generated as follows: First the data stream d(t) is filtered using a
Gaussian filter with the following impulse and frequency responses:
2)(
21
)( τt
eth−
= (2.1)
2)(
21
2)(τω
πτω−
⋅= eH (2.2)
This is similar to the familiar 2xe− shape of the Gaussian, or normal, probability
density function. In the equation, ω is the frequency in rad/sec and τ is a constant. A
peculiar property of the Gaussian filter is that both its frequency and impulse responses
are Gaussian. Figs. 2.3 and 2.4 show the impulse and frequency responses of a Gaussian
LPF, respectively. From equation (2.2), the bandwidth of the filter can be written as:
14
-2 -1.5 -1 -0.5 0 0.5 1 1.5 20
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Gau
ssia
n L
PF
im
pu
lse
resp
on
se
Time (µs)
21
−e
Fig. 2.3 Gaussian LPF impulse response
-2 -1.5 -1 -0.5 0 0.5 1 1.5 20
1
2
3
4
5
6
10-7
Gau
ssia
n L
PF
fre
que
ncy
res
po
nse
Frequency (MHz)
21)2ln(2 −
eπ
3dB
x
Fig. 2.4 Gaussian LPF frequency response
τπ1
2)2ln(
=B (2.3)
The Gaussian filter is often specified in terms of its relative bandwidth, relative to
the bit rate, or BT:
πτ 2
)2ln()()( bT
RateBitBandwidthFilterPeriodBitBandwidthFilterBT ==⋅= (2.4)
15
Higher BT means less intersymbol interference (ISI) but higher bandwidth. As a
good compromise, BT=0.5 is specified in Bluetooth standard [1]. So the bandwidth is
500kHz. Therefore, the output of the Gaussian filter is given by:
2)
)2ln((
21
)()( bTt
etdtgfπ
−
∗= (2.5)
Note that both )(td and )(tgf have a maximum and minimum of 1 and -1
representing 1 and 0 bits, respectively. The Gaussian filtered data )(tgf is then used to
frequency modulates the carrier. The output GFSK signal is expressed as:
))((2sin()( tgffftgfsk dC ⋅+= π (2.6)
Where Cf is the center frequency and df is the frequency deviation. Thus the
instantaneous frequency of the GFSK signal transitions between a frequency of dc ff − ,
representing binary 0, and dc ff + , representing binary 1. In Bluetooth standard [1], the
frequency modulation index is specified between 0.28 and 0.35. The modulation index is
the ratio between the frequency deviation 2 df and the bit rate. So the corresponding df
must be between 140 and 175 kHz. Fig. 2.5 shows representative examples of the
unfiltered, Gaussian filtered, and Gaussian FSK signal. For illustration purposes, the
GFSK signal in Fig. 2.5 has 1MHz center frequency.
2.3.3. Frequency Band
Bluetooth operates in the 2.4GHz Industrial Scientific Medicine (ISM) band. In a
vast majority of countries around the world the range of this frequency band is 2400MHz
– 2483.5MHz. By utilizing Time Division Duplex (TDD), transmitter and receiver share
the same frequency band. The regulators expect lots of devices to be using the same
spectrum, so they have set out rules for using ISM bandwidth to make sure that devices
16
can share the bandwidth. The rules state that you must spread the power of your
transmissions across the ISM band somehow. Two main methods are used for spreading
out the power: direct sequence spread spectrum (DSSS) and frequency-hopping spread
spectrum (FHSS). DSSS smears a transmission across a wide range of frequencies at low
power while FHSS spectrum uses a small bandwidth but changes (or hops) frequency
after each packet.
Bluetooth uses frequency-hopping spread spectrum as shown in Fig. 2.6. There are
79 channels of 1MHz each. Transmitters change frequencies 1,600 times every second.
Occasionally, two piconets may collide on the same channel, but they will just hop off to
new frequencies and retransmit any data that was lost. This technique also minimizes the
risk that portable phones or baby monitors will disrupt Bluetooth services, since the effect
of any interferer on a particular frequency will last only a tiny fraction of a second.
Bluetooth uses the master’s device ID to algorithmically determine the frequency
hopping pattern. This algorithm generates a unique pattern that is quite random and
exhibits an extremely long repeat cycle.
-1
0
1
Filt
ered
bas
band
gf(
t)
-1
0
1
Unf
ilter
ed b
asb
and
d(t)
0 5 10 15 20 25
-1
0
1
Gau
ssia
n F
SK
mod
ulat
ed g
fsk(
t)
10 -6 Fig. 2.5 GFSK modulation steps
17
Tim
Frequency
625µs
1MHz
Channel 79
Channel 1
Fig. 2.6 Frequency hopping diagram
2.3.4. Blocking Requirements
The blocking test for Bluetooth is performed by applying a Bluetooth-modulated
desired signal 10 dB (for Co-channel, 1 MHz and 2 MHz interference) or 3 dB (for all
other frequencies interference) over the reference sensitivity level. Then a Bluetooth-
modulated interfering signal is applied to the receiver at discrete increments of 1 MHz
from the desired signal with a magnitude as shown in Table 1. Five spurious response
frequencies are allowed at frequencies with a distance of equal or greater than 2MHz
from the wanted signal. On these spurious response frequencies a relaxed interference
requirement C/I = -17 dB shall be met, where C is the carrier power and I is the
interference power, respectively. Usually, each channel is allowed a different set of
exceptions. If a low-IF receiver is implemented, C/I degrades if I represents the image
signal. Spurious response frequency may be used to relax image rejection requirement in
Low-IF receivers.
2.3.5. Intermodulation Requirements
The adjacent channel immunity test is performed by applying one static sine wave
signal at f1 with a power level of –39 dBm and one Bluetooth modulated signal at f2 with
18
also a power level of –39 dBm to the input of the receiver while a wanted signal 6 dB
above the reference sensitivity is applied. The receiver must maintain a 0.1% BER. And
be aware when performing this intermodulation test, the effects of noise in receiver
channel are also there. It is necessary to make sure C/(I+N) is high enough to maintain
required BER, where N is the noise floor level.
2.3.6. Sensitivity
The actual sensitivity level is defined as the input level for which a raw bit error rate
(BER) of 0.1% is met. The requirement for a Bluetooth receiver is an actual sensitivity
level of -70dBm or better. The power level -70dBm is defined for 50Ω impedance. This
power can be written in terms of rms voltage and impedance as follows:
2
( ) 10log(power in ) 10log( ) 30rmsVP in dBm mWR
= = + (2.7)
So the power level -70dBm is equivalent to an rms voltage of 70.7µVrms. It is important
at this point to clarify the different representations of signal power:
Power in dBm ≡ decibels relative to one milliwatt.
Power in dBW ≡ decibels relative to one watt.
Power in dBV ≡ decibels relative to one volt.
Power in dBc ≡ decibels relative to carrier signal power.
2.4. What is 802.11b?
Also known as Wi-Fi (for Wireless Fidelity), 802.11b emerged in 1999 and is the
most popular wireless networking standard. Operating in the 2.4GHz radio band, 802.11b
is also the current mainstay of the 802.11 family of wireless networking standards
19
established by the IEEE (Institute of Electrical and Electronics Engineers). 802.11
defines the PHY (physical) and MAC (media access control) layers of the protocol. All of
the other layers are identical to the 802.3 (Ethernet) protocol. 802.11a was proposed
before 802.11b, hence the designation in 802.11a. 802.11b, however, came to the market
first.
802.11a/b uses the unlicensed spectrum for transmission and thus it must use spread
spectrum techniques. This process increases the communication channels interference
immunity or the processing gain, decreases interference between multiple users and
increases the ability to re-use the spectrum. 802.11b uses the 2.400 GHz to 2.483 GHz
spectrum. 802.11 is the wireless extension of 802.3 and supports all the underlying
protocols that Ethernet uses. An Access Point (AP) is the center of the Basic Service Set
(BSS) which may overlap partially, completely or not at all with each other without fear
of interfering with functionality. Mobile users can roam from AP to AP and these APs are
connected together with other APs using the same ESS_ID which forms an Extended
Service Set (ESS). Each AP has its own MAC and IP addresses and they are fault tolerant
when setup with multiple failure points. Addition of capacity to the network can be as
simple as adding APs to a new Ethernet port which uses the same ESS_ID.
802.11b uses DSSS (Direct Sequence Spread Spectrum) to disperse the data frame
signal over a relatively wide (30 MHz) portion of the 2.4 GHz band. This results in
greater immunity to radio frequency interference as compared to narrowband signaling.
Because of the relatively wide DSSS signal, you must set the 802.11b AP to specific
channels to avoid channel overlap which might cause a reduction in performance. In
order to actually spread the signal, the transmitter combines the Physical Layer
20
Convergence Procedure protocol data unit PLCP (PPDU) with a spreading sequence
through the use of a binary adder. PLCP is a frame modification technique used by
802.11b which is out of the scope of discussion in this paper. For higher data rates
(5Mbps, 11Mbps) 802.11b uses Complementary code keying (CCK) to provide spreading
sequences.
Detailed 802.11b standard specifications can be found in [3]. A summary of Wi-Fi
RF specifications is listed in Table 2.3.
Table 2.3 802.11b radio specifications
Frequency Band 2400 – 2483.5 MHz Duplex Time Division
Modulation DBPSK/DQPSK/CCK Channel Space/ number of channels 25 MHz / 3
Sensitivity (11Mbit/s) -76 dBm (for 0.08 FER ≡ 10-5 BER) Maximum Signal Level (11Mbit/s) -10 dBm Adjacent Channel
Rejection C/I25MHz -35 dB
Radio Frequency Tolerance ±60kHz
2.4.1. Wi-Fi Operation
Wi-Fi networks operate in two modes: ad hoc networks and infrastructure networks.
Ad hoc network is usually temporary. An ad hoc network is a self-contained group of
stations with no connection to a larger LAN or the Internet. It includes two or more
wireless stations with no access point or connection to the rest of the world. Ad hoc
networks are also called peer-to-peer networks and Independent Basic Service Sets
(IBSS). Fig. 2.7 shows a simple ad hoc network.
21
Infrastructure networks have one or more access points, almost always connected to
a wired network. Each wireless station exchanges messages and data with the access
point, which relays them to other nodes on the wireless network or the wired LAN. Any
network that requires a wired connection through an access point to a printer, a file server
or an Internet gateway is an infrastructure network. Fig. 2.8 shows an infrastructure
network [4].
Laptopcomputer
Laptopcomputer
Desktopcomputer
Fig. 2.7 An ad hoc wireless network with three stations using Wi-Fi
The Internet
Desktopcomputer
Desktopcomputer
LaptopcomputerLaptop
computer
Printer
Accesspoint
Networkbridge
Fig. 2.8 A simple infrastructure network using Wi-Fi
22
2.4.2. Modulation Format
802.11b is an extension of the 802.11 standard that uses two data rates, 1 and
2Mbits/s, use DBPSK and DQPSK modulation formats, respectively. In both data rates,
an 11-bit Barker sequence (+1, –1, +1, +1, –1, +1, +1, +1, –1, –1, –1) is used to spread
the signal at 11MHz chipping rate. In addition to these two rates, 802.11b provides 5.5
and 11Mbit/s data rates. 8-chip complementary code keying (CCK) is employed as the
modulation scheme at 11MHz chipping rate which is the same as the chipping rate in
802.11 standard, thus providing the same occupied channel bandwidth.
d1
d2
d3
d4
d5d6
d7
d8
φ1
φ2
φ3
φ4
Spreadingsequence C
8 QPSK chips/symbol
Q
I
time
2 3 4 3 41 2 4
2 3 34 2
( ) ( ) ( )
( ) ( )( ) ( )
, , ,
, , , ,1
j jj j
j jj j
C e e e e
e e e e
ϕ ϕ ϕ ϕ ϕϕ ϕ ϕ
ϕ ϕ ϕϕ ϕ
+ + + +
+
= ×
− − Fig. 2.9 Construction of the CCK modulated signal
CCK is a form of M-ary code word modulation where one from a set of M unique
signal code words is chosen for transmission [5]. The spreading function for CCK is
chosen from a set of M nearly orthogonal vectors by the data word. CCK uses one vector
from a set of 64 complex (QPSK) vectors for the symbol and thereby modulates 6-bits
(one-of-64) on each 8 chip spreading code symbol. Two additional bits are sent by QPSK
modulating the whole code symbol and this thus modulates 8-bits onto each symbol. Fig.
2.9 illustrates how the CCK modulated signal is constructed from the 8-bits data word.
23
The following formula is used to derive the CCK code words and is used for spreading
both the 5.5Mb/s and 11Mb/s:
1,,,,,,, )()()()()()()( 2332442434321 ϕϕϕϕϕϕϕϕϕϕϕϕϕ jjjjjjjj eeeeeeeeC −−×= +++++ (2.8)
Note that ϕ1 is added to all code chips, ϕ2 is added to all odd code chips, ϕ3 is added
to all odd pairs of code chips, and ϕ4 is added to all odd quads of code chips. The phases
ϕ1, ϕ2, ϕ3, ϕ4 are obtained in the 5.5Mb/s and 11Mb/s cases as follows:
2.4.2.1. CCK 5.5Mb/s Modulation
Four input bits, d0 – d3, are used to encode CCK phases ϕ1- ϕ4. The two bits d0 – d1
encode ϕ1 based on DQPSK as shown in table 2.4. The data di-bits d2 and d3 CCK
encode the basic symbol, as specified in Table 2.5. This table is derived from the formula
above by setting ϕ2= (d2 ×π) + π /2, ϕ3 = 0, and ϕ4 = d3 ×π.
Table 2.4 DQPSK encoding table
d0 d1 Even symbols phase change
Odd symbols phase change
00 0 π 01 π/2 -π/2 11 π 0 10 -π/2 π/2
Table 2.5 5.5Mb/s CCK encoding table
d2 d3 c1 c2 c3 c4 c5 c6 c7 c8 00 j 1 j -1 j 1 -j 1 01 -j -1 -j 1 j 1 -j 1 10 -j 1 -j -1 -j 1 j 1 11 j -1 j 1 -j 1 j 1
24
Table 2.6 QPSK encoding table
Di-bit pattern (di di+1) Phase00 001 π/2 10 π 11 -π/2
2.4.2.2. CCK 11Mb/s Modulation
At 11 Mb/s, 8 bits are transmitted per symbol. The first di-bit (d0 d1) encodes ϕ1
based on DQPSK as shown in table 2.4. The phase change of ϕ1 is relative to the phase
ϕ1 of the preceding symbol. The data dibits (d2, d3), (d4, d5), and (d6, d7) encode ϕ2, ϕ3,
and ϕ4, respectively, based on QPSK as specified in Table 2.6.
2.4.3. Operating Frequency Range
Wi-Fi operates in the same frequency range 2.4-2.4385GHz as Bluetooth.
2.4.4. Blocking Requirements
Adjacent channel rejection is defined between any two channels with >25 MHz
separation in each channel group. The adjacent channel rejection should be equal to or
better than 35 dB using
11 Mbit/s CCK modulation for both the desired and adjacent channels. The blocking
test is done at an input signal level 6 dB greater than the sensitivity level.
2.4.5. Intermodulation Requirements
There is no IM specified test for Wi-Fi. However the intermodulation requirement
can be derived from the Blocking test in section 2.3.2 due to the wide band interferer.
This will be discussed in details in section 5.4.
25
2.4.6. Sensitivity
The frame error rate (FER) is specified to be less than 0.08 at a frame length of 1024
octets (this is equivalent to about 10-5 BER for an input level of -76dBm. The FER is
specified for the 11Mbit/s CCK modulation.
2.5. Comparing Bluetooth and Wi-Fi
Both Bluetooth and Wi-Fi operate at the same ISM frequency band. However, there
are some major differences:
Speed: Bluetooth operates at a raw data rate 1Mbps, Wi-Fi at 11Mbps, a big speed
difference!
Applications: Bluetooth can be considered as a cable replacement technology. It is a
short-distance wireless technology having low cost and low power consumption. It is
intended to be a very simple technology in which devices can communicate with each
other without the need to configure hardware or drivers. Bluetooth is the choice for
connecting single devices when speed is not a major issue; it is best suited to low-
bandwidth applications such as sharing printers, synchronizing PDAs, using a cell phone
as a modem, and (eventually) connecting appliances to one another within a 30- to 60-
foot range. Wi-Fi technology on the other hand is really a wireless version of Ethernet.
Widespread popularity of Ethernet makes the Wi-Fi a very viable technology because it
can very easily be interfaced with any existing Ethernet setup or peripherals connected to
them.
Security: Bluetooth is probably a bit more secure than Wi-Fi. For one thing, Bluetooth is
designed to cover shorter distances than 802.11b. Also, Bluetooth offers two levels of
26
(optional) password protection. Wi-Fi has all the security risks associated with other
networks: Once someone has access to one part, he or she can access the rest.
Ease of use: A single Bluetooth device can be connected to up to seven other devices at
the same time. This makes it easy to find and connect to the device you are looking for or
to switch between devices, such as two printers. Wi-Fi is more complex; it requires the
same degree of network management as any comparable wired network.
Power: Bluetooth has a smaller power requirement than Wi-Fi, and devices can be
physically smaller, making it a good choice for consumer electronics devices.
Coexistence: Bluetooth and Wi-Fi share the same band of frequencies and could,
therefore, interfere with one another. Due to the modulation format, Bluetooth is more
likely to interfere with Wi-Fi than vice versa [6].
Spatial capacity: Wi-Fi has a rated operating range of 100 meters in free space. In a
circle with a 100-meter radius, three Wi-Fi systems can operate on a non-interfering
basis, each offering a peak over-the-air speed of 11Mbit/s. The total aggregate speed of
33Mbit/s, divided by the area of the circle, yields a spatial capacity of approximately
1kbit/s per square meter. Bluetooth, on the other hand, with its low power mod has a
rated 10-meter range and a peak over the air speed of 1Mbit/s. At least 10 Bluetooth
piconets can operate simultaneously in the same 10-meter circle with minimal
degradation, yielding an aggregate speed of 10Mbit/s. Dividing this speed by the area of
the circle produces a spatial capacity of approximately 30kbit/s per square meter.
Bluetooth versus Wi-Fi RF specifications: Here are the main differences between the
RF specifications of the two wireless standards.
27
• Operating frequency range: Bluetooth and Wi-Fi operate at the same ISM
frequency band 2.4-2.483GHz.
• Channel Bandwidth: Wi-Fi bandwidth (22MHz) is much wider than Bluetooth
(1MHz)
• Data rate: Wi-Fi supports 4 different data rates (1,2,5.5, 11Mbit/s) while
Bluetooth has only one data rate (1Mbit/s).
• Modulation format: Bluetooth uses GFSK modulation while Wi-Fi uses
DBPSK/DQPSK/CCK depending on the data rate.
• Sensitivity: Bluetooth sensitivity is -70dBm while Wi-Fi specifies a lower
sensitivity level at -76dBm.
• Adjacent Channel Rejection (ACR): Bluetooth specifies ACR > 40dB at 3MHz
offset while Wi-Fi has ACR > 35dB at first adjacent channel (25MHz offset).
2.6. What is UWB?
Traditional wireless systems operate within the confines of a narrow band of
frequencies assigned by the government regulatory authorities. Ultrawideband is
different. UWB technologies occupy a broad swath of frequencies, typically 1.5 to 4GHz
wide, that covers many already assigned frequency bands in the 1 to 6 GHz range. UWB
purports to occupy these frequencies without causing undue interference. It does so by
emitting a power so low that it meets US Federal Communication Commission (FCC)
constraints. FCC Part 15, set for incidental radiation from devices like laptops, hair
dryers, and electric drills. However, UWB systems need a waiver from the FCC Part 15
rules because they function as intentional radiators. The FCC has published a notice of
the proposed rule making that could lead to such a waiver.
28
Traditional wireless systems operate within the confines of a narrow band of
frequencies assigned by government regulatory authorities. Ultrawideband is different
[7], UWB technologies occupy a broad swath of frequencies, typically 1.5 to 4 GHz
wide, that cover many already-assigned frequency bands in the 1 to 6 GHz range. UWB
purports to occupy these frequencies without causing undue interference. It does so by
emitting a power so low that it meets US Federal Communication Commission
constraints, FCC Part 15, set for incidental radiation from devices like laptops, hair
dryers, and electric drills. However, UWB systems need a waiver from the FCC Part 15
rules because they function as intentional radiators. The FCC has published a Notice of
Proposed Rule Making [8] that could lead to such a waiver. Because of their very low
radiated power, UWB systems are impractical for long-range communication use, but
they appear ideal for SRW applications, particularly in the wireless PAN range of 10
meters or less. Laboratory systems have already demonstrated data bandwidths in excess
of 100 Mbps over distances greater than 10 meters, with less than 200 microwatts of
average radiated power—about one fifth that of a low-power Bluetooth link. Technically,
a UWB system is defined as any radio system that has a bandwidth greater than 25
percent of its center frequency, or greater than 1.5 GHz. UWB technology first appeared
in the 1980s, primarily for use in radar [9]. Recent advances in low-cost, low-power
switching technology and processing have made it practical to consider using UWB for
consumer-grade communication devices. UWB systems emit very narrow pulses with
sharp rise times, with the narrowness of these pulses giving rise to UWB’s broadbanded
nature. Systems based on this emerging technology vary widely in their projected spatial
capacity, but one UWB developer has measured peak speeds of more than 50 Mbps at a
29
range of 10 meters. That developer projects that at least six such systems could operate
within the same 10-meter-radius circle and experience only minimal degradation [10].
Following the same calculation process, the projected spatial capacity for such a system
would be more than 1,000 Kbps per square meter. Fig. 2.10 provides a side-by-side
comparison of Bluetooth, Wi-Fi, and UWB spatial capacities. As shown, UWB appears
to offer a substantial advantage. The Hartley-Shannon law offers a plausible reason for
UWB’s spatial-capacity advantage, as described in the following equation:
)1(log2 NSBC += (2.9)
Where:
C = Maximum channel capacity, in bits per second
B = Channel Bandwidth in Hertz
S = Signal power, in Watts
N = Noise power, in Watts
Because the upper bound on a channel’s capacity grows linearly with the total
available bandwidth, UWB systems, which occupy 1.5 GHz or more, have inherently
greater headroom for expansion than more bandwidth-constrained systems. UWB
technology for SRW communications is still in its earliest days. It is not yet standardized,
has its own multiple competing variations, and has not received necessary regulatory
approvals. Nonetheless, as a long-term target, UWB appears to have enormous potential,
especially as a wireless PAN technology.
30
0
100
200
300
400
500
600
700
800
900
1000
Bluetooth(1kbps/m2)
802.11b(30kbps/m2)
UWB(1,000kbps/m2)
Spat
ial c
apac
ity (k
bps/
m2 )
Fig. 2.10 Comparison of spatial density of Bluetooth, Wi-Fi, and UWB
31
CHAPTER III
BLUETOOTH RECEIVER ARCHITECTURE AND CHANNEL
SELECTION FILTER
Our Bluetooth receiver design project started in the summer of 2001 with a group of
6 PhD students. We started by studying different receiver architectures [11] to achieve
the highest level of integration, lowest power consumption, and best performance. Of
course, all these requirements are not met in a single architecture, and therefore, tradeoffs
were closely studied to find the best architecture that meets the standard specifications
with enough margins at lower cost. Two very common architectures are used for
Bluetooth receivers, direct-conversion and low-IF.
Although the direct-conversion architecture lends itself to higher integration levels
and lower power consumption, it is plagued by quadrature demodulation phase errors,
quadrature gain phase mismatch, DC offsets, 1/f noise, and LO feedthrough [12]. Low-IF
architecture [13] can be used to avoid the DC offset and 1/f complications associated with
direct-conversion. However, Low-IF architecture suffers from the image problem due to
the non-zero IF frequency.
The choice of the most suitable receiver architecture depends on many parameters in
the wireless standard (e.g. channel bandwidth, preamble time, blocking specifications,
sensitivity, modulation format, etc..). In the following two sections, both possible
architectures of the Bluetooth receiver will be discussed in some detail. The channel
selection filter design, which I was responsible for, is described in detail in section 3.3.
The receiver and the filter were fabricated in TSMC 0.35µm CMOS process.
32
3.1. Direct-Conversion Receiver Architecture
In direct-conversion receiver (DCR) architecture, the signal is down-converted
directly from RF to baseband. A low-pass filter is then employed to suppress nearby
interferers as shown in the simplified diagram in Fig. 3.1. The use of quadrature I and Q
channels is necessary in the Bluetooth case because the signal is frequency-modulated,
and therefore the two sidebands of the RF spectrum will carry different information. The
spectrum of the complex output signal I+jQ will be a replica of the signal spectrum at RF,
but down-converted around dc. Despite the simplicity of the DCR architecture, it suffers
from some serious design issues that do not exist or are not as serious in low-IF receivers.
LO
LNA90o
ADC
ADC
LPF VGA
Fig. 3.1 Direct-conversion receiver architecture
3.1.1. DC Offsets
There are different sources of DC offsets in an integrated receiver; (1) components’
mismatches, (2) LO self-mixing, and (3) interferers self-mixing. These sources are
explained as follows:
(1) Typical MOS transistor VT mismatches are in the order of few millivolts. This
might be quite higher than the desired signal level at the mixer output.
33
(2) Due to capacitive and substrate coupling, isolation between the LO port and the
inputs of the mixer and the LNA is finite. This effect is called LO leakage. The
leakage signal appearing at the input of the LNA and the mixer is now mixed with
the LO signal thus producing a dc component at the mixer output. If the LO signal
level is 0dBm, and isolation between LO and LNA input is 60dB, then the LO
signal at the LNA input is about -60dBm, quite substantial compared to the
minimum signal (sensitivity) level at the receiver input.
(3) This effect is similar to LO self mixing. When a large interferer leaks from the
LNA or mixer input to the mixer LO port, it mixes with itself and generates a low
frequency beat at the mixer output corresponding to amplitude variations in the
interferer. This resulting offset is even harder to reject since it is varying with
time.
This means that if the desired signal level at the end of the receiver chain is at the
full swing of the final stage, the dc offset generated by mismatches will saturate the
receiver stages. Therefore, this dc offset has to be rejected before it gets amplified by the
receiver stages. A possible approach to removing the offset is to employ ac coupling, i.e.
high-pass filtering, in the down-converted signal path. However, since the spectrum of
the Bluetooth GFSK signal exhibits a peak at dc, such signal may be corrupted if filtered
with high cutoff frequency.
3.1.2. 1/f Noise
In modern technologies and for the minimum gate-length transistors required by RF
circuits, the 1/f noise (also called flicker noise) component might exceed the white noise
up to several megahertz. On the one hand, flicker noise is not a limiting effect for linear
34
RF circuits, as in the low noise amplifier (LNA) since the operating frequency is much
higher than the corner frequency. On the other hand, since minimum length transistors
are used in the switching transistors and due to the nonlinear operation of the mixer and
the finite slope of the LO signal, flicker noise of the switches appears at the baseband
output of the mixer. Flicker noise of the transistors used in the baseband circuits also falls
in the signal band and degrades the system noise figure (NF). NF degradation depends on
the flicker noise corner frequency and the channel bandwidth. In the case of Bluetooth,
the -3dB bandwidth of the signal is about 500kHz, while the 1/f corner frequency is about
1MHz or even larger for smaller transistor lengths. This NF degradation might be so
significant that it disqualifies DCR architecture as the optimum choice for Bluetooth. The
effect of flicker noise can be reduced by a combination of techniques. As the stages
following the mixer operate at relatively low frequencies, they can incorporate longer
devices to minimize the magnitude of the flicker noise. Moreover, periodic offset
cancellation also suppresses low-frequency noise components through correlated double
sampling.
3.1.3. Even Order Distortion
Unlike other architectures, even-order distortion in the LNA and mixer input
transistor becomes problematic in DCR architecture. Suppose, as shown in Fig. 3.2, two
strong interferers ( )cos()cos( 2211 tAtA ωω + ) close to the desired channel experience a
second order nonlinearity in the LNA represented as )()()( 221 txtxty αα += , then y(t)
contains a low frequency term tAA )cos( 21212 ωωα − . Upon multiplication by )cos( tLOω
in an ideal mixer, such a term is translated to high frequencies and hence becomes
unimportant. In reality, however, mixers exhibit a finite direct feedthrough from the RF
35
input to the IF output due to mismatches between transistors and deviation of LO duty
cycle from 50%. The natural solution to suppress even-order distortion is to use
differential LNA and mixer. However, two issues arise here. First the antenna and the
duplexer filter are usually single-ended. This necessitates the use of a balun (transformer)
to do the single-ended to differential conversion. Baluns typically exhibit several decibels
of loss at high frequencies. This loss directly raises the overall system noise figure.
Second, differential LNA requires more power consumption than the single ended
counterpart to achieve the same noise figure.
ωω2ω1
ωω2ω10 ∆ω
∆ω0
InterferersDesiredsignal
ω
cos(ωLO)
Fig. 3.2 Effect of even order distortion
3.1.4. I/Q Mismatch
Phase and magnitude mismatches between I and Q branches corrupts the
downconverted signal. However, in the case of Bluetooth, since the modulation format in
binary GFSK, I/Q mismatch is not a serious problem. Therefore, I/Q mismatch is much
36
less troublesome in DCR than in low-IF architecture where I/Q mismatch results in finite
image rejection (this effect will be discussed in the next section).
3.2. Low-IF Receiver Architecture
IF receivers have been in use for a long time, and their principle of operation is very
well known [14]. In an IF receiver, the wanted signal is down-converted to from its
carrier to the IF by multiplying it with a single sinusoidal signal as shown in Fig. 3.3. The
main disadvantage here is that apart from the wanted signal, an unwanted signal at a
frequency called the image frequency (which is IFf2 away from the wanted frequency) is
down-converted to the same IF frequency. To avoid corrupting the wanted signal, the
image signal must be suppressed before down-conversion by means of a band-pass RF
filter. The Q of such filter is proportional to IFRF ff / . High Q external SAW or ceramic
filters (typically 50 or more) are used for this purpose. Such filters are bulky and require
impedance matching at input and output, which usually raises the power consumption
that is needed in order to drive this low impedance. Furthermore, the IF frequency cannot
be made arbitrarily small due to the limited Q of the external filter, and hence, raising the
power consumption of the circuits operating at the IF frequency.
The image frequency problem can be mathematically explained as follows. When RF
signal is multiplied by a single sinusoidal signal )cos( LOω , it is equivalently multiplied
by two exponentials LOje ω− and LOje ω . Considering the signal diagram in Fig. 3.4, the
spectrum of the down-converted signal is constructed by shifting the RF spectrum to the
left and to the right by LOω and add the two shifted replicas. The result is two
overlapping spectra, of the signal and the image, at the IF frequency.
37
LO
LNA
BPF
ωωLO+ωIFωLO-ωIF ωLO ωωIF0
cos(ωLO)RF
filter
ωωLO+ωIF
Imagesignal
Wantedsignal
ωLO-ωIF ωLO
−ωIF
Fig. 3.3 IF receiver architecture
WantedsignalImage
signal
0 ωLO+ωIFωLO-ωIF ωLO−ωLO
LOsignal
PSDBefore
down-conversion
0
PSDAfter
down-conversion
ωIF−ωIF
LOje ω−LOje ω
shift byshift by
ω
ω
Fig. 3.4 Down-conversion with a single sinusoidal signal
An obvious solution to avoid this problem is to multiply the RF signal by only
exponentials, say LOje ω− , which means that the down-converted signal is simply a single
38
shifted replica of the RF signal and therefore, no overlapping of signals spectra. The only
problem now is that the signal LOje ω− is a complex signal with real part ( )cos( LOω ) and
imaginary part ( )sin( LOω− ). To implement this complex multiplication using real
components, two signal branches I and Q must be constructed. In the I (in-phase or real)
branch, the RF signal is multiplied by )cos( LOω , while in the Q branch (quadrature-phase
or imaginary), the RF signal is multiplied by )sin( LOω− . Fig. 3.5 shows the signal
diagram of the complex down-conversion operation. It is important to note that in each
branch, the down-converted signal contains both the wanted and the image signals at the
same IF frequency. However, the complex signal OO jQI + has the wanted signal at IFω
and the image signal at IFω− . The image signal can then be rejected by means of a
complex filter, which will be described in detail in the following section. The same filter
is also used for channel selectivity. Therefore, the Q of such filter is proportional to
BWIF /ω (BW is the channel bandwidth) which is small for low IF frequencies. This is
the basic idea behind low-IF receiver architecture.
WantedsignalImage
signal
0 ωLO+ωIFωLO-ωIF ωLO−ωLO
LOsignal
PSDBefore
down-conversion
0
PSDAfter
down-conversion
ωIF−ωIF
shift by LOje ω−
ω
ω
Fig. 3.5 Down-conversion with a single exponential
39
Fig. 3.6 shows the basic low-IF receiver architecture. Although low-IF receiver
avoids the problems that exist in DCR and high-IF architectures, it has some design
issues. Namely, its image rejection capability is limited by matching between I and Q
branches and between the quadrature LO outputs. The effect of these mismatches is
studied in the following section.
LO
LNA90o
ωωIF0−ωIF
ω
j
)cos( LOω
)sin( LOω−
-j -j
j
ωωLO+ωIFωLO-ωIF ωLO
Imagesignal
Wantedsignal
Fig. 3.6 Basic low-IF receiver architecture
3.3. Bluetooth Receiver Architecture
In a baseband Bluetooth signal, 99% of the signal power is contained within the DC
to 430kHz bandwidth. Therefore, if direct-conversion architecture is used, the flicker
noise and DC offset might significantly degrade the signal-to-noise ratio (SNR). Hence, a
low-IF architecture seems to be a suitable architecture in Bluetooth, especially when
considering the relaxed image rejection requirement in the Bluetooth standard [1]. To
relax the image rejection requirement and reduce the folded-back interference level, a
40
very low-IF is preferable, i.e. half of the channel bandwidth. However, such a very low-
IF requires a sharp cut off from the channel selection filter to reject the DC offset and
flicker noise. On the other hand, a higher IF improves the demodulator performance, but
the required selectivity of the channel selection filter will increase, and power
consumption will be higher. As a good compromise, an IF of two times the channel
bandwidth is chosen, i.e. 2 MHz. For a low-IF Bluetooth receiver, the image signal is an
in-band Bluetooth modulated adjacent channel interference, which becomes co-channel
interference after frequency down-conversion. It has been verified that an image rejection
of 33 dB is sufficient to meet the Bluetooth specifications [15]. For the on-chip image
rejection, there are several potential architectures: Hartley architecture, Weaver
architecture [16], passive RC polyphase filter and active polyphase filter (also called
complex filter). For the Hartley architecture, the high channel bandwidth to IF ratio
makes the design of the 90o phase shifter very difficult. Weaver architecture requires an
extra set of mixers, a frequency synthesizer and high order band-pass filters to reject the
second image; thus, the power consumption and silicon area penalty is high. Polyphase
filter (also called complex filter) can be used in front of the ADC or it can be embedded
in a Σ∆ ADC loop [17].Although passive RC polyphase filters can achieve a high image
rejection ratio [18], they cannot achieve the required attenuation of the adjacent channel
interference, especially those strong folded-back interferences, due to their limited
selectivity. Extra filtering is then required to reject the adjacent channel interference,
which is also true for Hartley and Weaver architectures. Another drawback of a passive
RC polyphase filter is that the finite input impedance loads the RF mixers. Fortunately,
an active complex filter can achieve good image rejection and adjacent channel
41
interference rejection. Fig. 3.7 shows one embodiment of the low-IF receiver [15]. The
RF signal is amplified and down-converted to IF by the RF front end, then, the channel
selection is performed by an active complex filter, which is described and proposed in
this paper, and next the IF signal is passed through an amplitude limiter which removes
any amplitude perturbations. As a final stage, a GFSK demodulator is employed [19].
The Bluetooth standard allows a transmitted center frequency offset as large as ±100
kHz in one time slot [1]. If the frequency offset cannot be cancelled before the channel-
selection filter, the passband of the filter has to be extended to pass the desired signal
with frequency offsets up to 100kHz. Since 99% of the baseband signal power is
contained within 430kHz band, the complex filter passband becomes
( ) kHz10601004302 =+ centered at 2MHz.
LNA 90o
PLL
RFFilter
RSSI
Synthesizerand VCO
ComplexFilter
Data_outGFSKDemodulator
LimiterMixer
I
Q
Fig. 3.7 A Low-IF Bluetooth receiver architecture [15]
Complex filter designs were found in the literature [20-23]. However, these filters
either consume significant power and area (e.g. in [21], the filter draws 90mA from 5V
supply and occupies an area 7.5mm2), or use a well-controlled special analog process
(e.g. in [20], the process parameters are controlled within 1%). Other complex filters are
42
reported as part of the receiver design, and therefore, details about the filter performance
were not given [23].
Partial results of the proposed complex filter have been reported in [15]. In this
work, a pseudo differential OTA-C complex filter is designed and tested. A pseudo
differential OTA is used to improve filter linearity and device area. A non-conventional
frequency tuning scheme, simpler than conventional frequency tuning techniques, is used
to compensate for process variations in a mainstream low cost TSMC 0.35µm CMOS
process. This is a step forward, towards a highly integrated, low-cost Bluetooth receiver.
3.4. Channel Select Complex Filter Design
3.4.1. Complex Filter Theory
Complex filters are not new; they were invented by Sedra et al [24] in 1985.
However, to justify how to implement them, a brief theoretical discussion follows. To
understand the ability of complex filters to reject the image signal, consider the complex
representation of the receiver block diagram shown in Fig. 3.8. For the sake of
illustration, we will assume that only the desired signal and its image are present at the
mixer input. Without loss of generality, we will assume the signal and the image
frequencies are IFLO ωω + and IFLO ωω − , respectively. After eliminating the double LO
frequency term by the mixer output low-frequency pole, the result of mixing the LO and
RF signals in the complex domain is:
QItj
imagetj
sigmixer jBBexexGB IFIF +=+= − )( ωω (3.1)
43
Where BI and BQ are the real and imaginary parts of the mixer output and can be
expressed1 as:
))sin()sin((
))cos()cos((
txtxGB
txtxGB
IFimageIFsigmixerQ
IFimageIFsigmixerI
ωω
ωω
−=
+= (3.2)
Complex FilterH(s) = LP(s-jωIF)X
tjsigmixer
IFexGC ω=
QI
tjimage
tjsigmixer
jBBexexGB IFIF
+=
+= − )( ωω
)sin()cos( tjte
LOLO
tj LO
ωω
ω
+=
txtxA
IFLOimage
IFLOsig
)cos()cos(ωω
ωω
−+
+=
Fig. 3.8 Receiver image rejection architecture in the complex domain
Note that in the above equation the desired (image) signal in the I branch leads (lags)
the Q branch by 90º. Fig. 3.9 illustrates the complex mixing operation on the desired
signal and its image. Note that after down-conversion, the 2ωIF frequency separation
between the signal and the image is still preserved. The complex channel select filter [24]
is then a frequency-shifted version of a low pass filter response. This means that the filter
can pass the signal at ω=ωIF, while attenuating the signal at ω=-ωIF. Since the filter has
unsymmetrical frequency response around the jω axis, its time domain response is
complex2 (here comes the name complex filter). However, the complex filter frequency
response is symmetrical around the ωIF. Since the blocking specifications of a receiver
are symmetrical around the desired signal frequency, this is considered an advantage of
complex filter over real BPF that has unsymmetrical frequency response around its center
frequency. 1 Note that the notations I and Q in Fig. 3.1 correspond to BI and BQ, respectively in Fig. 3.2. 2 In real filters, complex poles are always conjugate, but in complex filters one single complex pole is possible.
44
These complex operations are practically performed as follows. Multiplication of the
real RF signal by LOje ω is practically performed using a quadrature mixer, which
basically consists of two mixers whose LO inputs are in quadrature phase, as shown in
Fig. 3.10. In the complex signal representation in Fig. 3.8, the desired signal at the mixer
output is located at the positive IF frequency while the image signal is located at the
negative IF frequency. In the real implementation in Fig. 3.10, the desired (image) signal
in the I branch leads (lags) the Q branch by 90°. Phase and gain imbalances at the mixer
output, due to LO and mixer mismatches, will cause the image signal at -ωIF to spill over
the image band at ωIF. As a result the image rejection ratio (IRR) will be limited by these
mismatches. It can be shown that the rejection limit (in dB) is given by:
))2
()2
(log(sin10)))2
(cos)2
()2
(log(sin10(indB)IRR 22222max
∆+≈
∆+=
θθθ (3.3)
Where ∆ and θ are the gain and phase imbalances, respectively. For example, for
IRR > 30dB, the maximum tolerable phase and gain mismatches (assuming equal
contribution to IRRmax) are 2.5° and 4.2%, respectively. See [25] also about mismatch
effects on complex filters.
The complex filter, in turn, is able to make the distinction between the signal and the
image based on the phase difference between the I and Q branches. In the complex
domain, the complex BPF is a frequency-shifted version of an LPF. To convert an
arbitrary LPF to a complex BPF centered at ωIF, every frequency dependent element in
the LPF [24] should be altered to be a function of s- jωIF instead of s. The basic
frequency dependent element in a filter is the integrator. Consider the simple case of
converting a first order LPF with cut-off frequency ωLP, to a complex filter BPF centered
45
at ωIF. The LPF response is shifted in frequency by placing it in a complex feedback loop
as shown in Fig. 3.11(a). The complex input-output relation is given by:
iIFLP
oo x
jsx
ωωω
−+= (3.4)
where iQiIi jxxx += and oQoIo jxxx += , then from (3.4):
)( oQo
IFiI
LP
ooI xx
sx
ωω
ωω
−+
= (3.5.a)
)( oIo
IFiQ
LP
ooQ xx
sx
ωω
ωω
++
= (3.5.b)
IFLO ωω − IFLO ωω +LOω
IFω2Rxinput
ω
imagesignal
(a)
IFω− IFω
IFω2
Aftercomplexmixing
ω
complexfilter
response
(b)
Fig. 3.9 Frequency translation of a complex (quadrature) mixer (a) before complex mixing (signal A in Fig. 3.8) (b) after complex mixing (signal B in Fig. 3.8)
46
)cos( tLOω
))cos(
)cos((
tx
txGB
IFimage
IFsigmixerI
ω
ω
+
=
Complex Filter
H(s)=LP(s-jωIF)
)cos( IFsigmixerI xGC ω=
)sin( tLOω
))sin()sin((
txtxGB
IFimage
IFsigmixerQ
ω
ω
−
= )sin( IFsigmixerQ xGC ω=tx
txA
IFLOimage
IFLOsig
)cos(
)cos(
ωω
ωω
−+
+=)cos( tLOω
))cos(
)cos((
tx
txGB
IFimage
IFsigmixerI
ω
ω
+
=
Complex Filter
H(s)=LP(s-jωIF)
)cos( IFsigmixerI xGC ω=
)sin( tLOω
))sin()sin((
txtxGB
IFimage
IFsigmixerQ
ω
ω
−
= )sin( IFsigmixerQ xGC ω=tx
txA
IFLOimage
IFLOsig
)cos(
)cos(
ωω
ωω
−+
+=
Fig. 3.10 Practical implementation of the receiver image rejection
Equation (3.5) is implemented as shown in Fig. 3.11(b). For illustration purposes, an
active-RC implementation of this first order complex filter is shown in Fig. 3.11(c). Note
that an inverting amplifier is needed in the cross feedback from Q branch to I branch. If a
differential implementation is used, this extra inverting amplifier can be avoided by
exchanging the differential signals. The corresponding pole locus of the prototype and the
complex BPF are shown in Fig. 3.12. For the special case of an integrator prototype,
when ωLP = 0, the above transformation still holds and the integrator response will be
shifted to ωIF. If the above transformation is applied to every frequency dependent
element in the LPF prototype, the entire LPF frequency response will be shifted to ωIF.
Fully differential active-RC implementation of the first order complex filter is illustrated
in Fig. 3.13. Note that the inverting amplifier is eliminated in this case.
3.4.2. Complex Filter Implementation
3.4.2.1. Filter Approximation
In OTA-C filters, the lowpass to complex filter transformation is done by replacing
each pair of integrators, in I and Q branches, by the circuit shown in Fig. 3.14 for an
OTA-C filter. Differential OTA-C implementation of the linear frequency translation is
depicted in Fig. 3.15. System level simulations show that a complex filter based on a 4th
47
order Chebychev LPF or 6th order Butterworth LPF may be sufficient to achieve the
required selectivity. The Butterworth approximation is preferred for two reasons. First, it
has small group delay variation (0.6 µs) within the pass band. Second, since all the poles
have the same angular frequency in a Butterworth filter, the cross-coupled OTA’s will
have the same transconductance value in the entire filter resulting in better matching
between filter stages and between the filter and the frequency tuning circuit. The highest
quality factor (Qp) in the LPF prototype is 2. This small Qp can tolerate process
mismatches without affecting the filter performance significantly, thus it can be realized
easily without using Qp tuning. However, a frequency tuning circuit is required to
compensate for the expected ωo variations due to the process technology variations. To
simplify the LPF to bandpass filter (BPF) transformation, the LPF prototype should have
only grounded capacitors. If the LPF prototype was to include floating capacitors, two
pairs of cross-coupled OTA’s would be used for each floating capacitor (one cross-
coupled pair for each capacitor terminal) for frequency shifting. Therefore, floating
capacitors in the LPF prototype means increased area and power of the complex filter.
The LPF prototype is implemented using three biquads.
3.4.2.2. OTA Topology and CMFB
In order to reduce the input referred noise, the least number of transistors is used in
the OTA, as shown in Fig. 3.16. This OTA was preferred over Nauta’s transconductor
[26] for two reasons. First, Nauta’s transconductor has some circuitry to ensure CM
stability which may not be needed in some cases, and therefore, may consume more
power. Second, Nauta’s transconductor is tuned through its supply voltage, and therefore,
a buffer with high current driving capability and low output resistance is needed to drive
48
the OTA supply node. The fact that both NMOS and PMOS transistors contribute to the
transconductance in Nauta’s transconductor is not really relevant in this case since a low
transconductance value is used.
LP
o
s ωω+
o
IFjωω
xi xoIFLP
o
js ωωω
−+xi xo
LP
o
s ωω+
o
IFjωω
xi xoIFLP
o
js ωωω
−+xi xo
(a)
LP
o
s ωω+
o
IF
ωω
Σ
Σ
IF
o
ωω
−
LP
o
s ωω+iQx
iIx
oQx
oIx
_
+
xiI-xoI
RO
CLP
RLP
_
+
xiQ-xoQ
RO
CLP
RLP
RIF
RIF -1
(b) (c)
Fig. 3.11 LPF shifted to ωIF (a) conceptual complex representation (b) actual building block implementation (c) Active-RC implementation
49
Imaginary
Real− ωLP
ωIF
LPF prototype pole
AfterFrequency transformation
Imaginary
Real− ωLP
ωIF
LPF prototype pole
AfterFrequency transformation
Fig. 3.12 Pole locus of the LPF prototype and the complex BPF
xoQ+
RLP
xiQ+RO
CLP
xiQ-RO
CLP
RLP
xoQ-
xoI+
RLP
xiI+RO
CLP
xiI-RO
CLP
RLP
xoI-
RIF
RIF
RIF
RIF
Fig. 3.13 Fully differential active-RC complex first-order LPF
50
C
C
ωoC
ωoC
ωoC
ωoC
C
C
xiI
xiQ
xoI
xoQ
xiI
xiQ
ωIFC ωIFC
xoI
xoQ
Fig. 3.14 Linear frequency translation to convert LPF to complex BPF
ωoC
ωIFC
C
C
ωoC
C
C
xiI+
xiI-
xoI+
xoI-
xiI+
xiI-
xoI+
xoI-
ωoC
C
C
ωoC
C
C
xiQ+
xiQ-
xoQ+
xoQ-
xiQ+
xiQ-
xoQ+
xoQ-
ωIFC
Fig. 3.15 Differential OTA-C implementation of linear frequency translation
51
VDD
vi-M1 M2
M3 M4Vb
io+io-
vi+
VDD
vi-M1 M2
M3 M4Vb
io+io-
vi+
Fig. 3.16 Pseudo differential OTA
Long channel transistors (6 µm) are used in the OTA in Fig. 3.16 to enhance the
output resistance, improve matching, and reduce flicker noise. A pseudo differential
architecture is used to reduce the required supply voltage. The signal swing is determined
by the overdrive voltage (Vdsatn) of the input transistors M1 and M2. The minimum and
maximum acceptable single-ended input levels to the OTA are VTn and VTn+2Vdsatn,
respectively. Therefore, the minimum power supply voltage is VTn + 2Vdsatn + Vdsatp. The
transconductance of the OTA is linearly proportional to the input common mode (CM)
voltage VCM:
)('TnCMnm VV
LWKg −= (3.6)
If Vb is a fixed bias voltage (i.e. no CM control), the common mode rejection ratio
(CMRR) of the OTA equals unity. Therefore, Vb must be controlled using the CM input
or output signal of the OTA to improve its CMRR.
52
3.4.2.3. CMFF and CMFB Interconnection Strategy
Here we discuss an alternative approach to avoid the use of conventional CMFB
circuits which often can be area and power hungry. Fig. 3.17(a) shows the CM equivalent
of an OTA with no CM control. Note that input CM signal is transferred to the output
through the CM transconductance, which happens to be the same value as given in (6).
Unless the CM impedance at the output node is low enough, this biasing approach
provides high CM voltage gain and may cause CM instability. The biasing circuit of the
OTA in Fig. 3.17(a) is shown in Fig. 3.18(a). VCM in the biasing circuit is a fixed bias
voltage and is not dependent on CM input or output signals of the OTA.
gm
VCM
vi+
vi-
vo+vo-
Vb
viCMvoCM
gm viCM
OTA without CM control CM equivalent circuit
Biasingcircuit
(a)
gm
CMDVCM
vi+
vi-
vo+vo-
NFB
viCMvoCM
cmfbg1gm viCM
CMFB CM equivalent circuit
Auxiliarycircuit
(b)
Fig. 3.17 (a) No CM control (b) CMFB circuit (c) CMFF circuit
53
gm
vi+
vi-
vo+vo-
NFF
viCMvoCM
CMFF CM equivalent circuit
CMD
(c)
Fig. 3.17 Continued
M13
M14
VDD
VCM
Vb
to OTA in Fig. 3.15(a)
M5 M6
M7M8
M9
M10
M11
M12
NFBCMFB output
(non-inverting)
NFFCMFF output
(inverting)
VDD
VCMv- v-
CMD circuit Auxiliary circuit
to OTA inFig. 3.15(c)
(to OTA inFig. 3.15(b)
(a) (b)
Fig. 3.18 (a) Biasing circuit (b) The CMD and auxiliary circuit required for CMFF and CMFB
54
gm
gmQ1
gmiiI 1 2
Vbfrom Fig. 10(b)
OTA1 OTA2
OTA3
gm ioI
CMD2
From OTA5 in Qchannel
OTA4
gmβ1
CMD1
From OTA6 in Qchannel
gmβ1
To node in Qchannel
OTA5OTA6
LP biquadprototype
Linearfrequency
transformation
NFF
NFB
NFF
2To node in Qchannel
1
To nextbiquad
VCM
Auxiliarycircuit
(a)
LP biquadprototype
1 2
gmβ1gmβ
1gmβ1 gmβ
1
iiI ioI
LP biquadprototype
iiQ ioQ
1 2
(b)
Fig. 3.19 (a) I branch of the complex biquadratic section (b) Conceptual complex biquad
55
To enhance the CMRR of the OTA’s, a CMFB or a common-mode feed forward
(CMFF) is used. If the output CM impedance is high, then CMFB is needed to lower this
impedance and to fix the DC operating point. This is illustrated in Fig. 3.17(b) where the
output CM impedance becomes 1/gmfb, where gmfb is the transconductance of the CMFB
loop. In Fig. 3.17(b), the common mode detector (CMD) senses the CM signal at the
output node and feeds the correction signal to the bias voltage Vb of the OTA. To ensure
CM stability, the magnitude of the closed loop CM gain (GCM) should be less than one.
For a load resistance of RL, the GCM gain of the circuit if Fig. 3.17(b), when the OTA of
Fig. 3.16 is used, can be expressed as:
Lmp
LmnCM Rg
RgGα+
−=
1 (3.7)
Where gmn and gmp are the transconductances M1 and M3, respectively, in Fig. 3.15,
and α is the voltage gain of the CMD. Since the CMFB is used in cases when the load
resistance is high, we can assume that 1mp Lg Rα >> . In this case, the CM stability
condition reduces to the following condition:
mnmp gg >α (3.8)
On the other hand, if the output CM impedance is sufficiently small, CMFB is not
needed and CMFF is used to isolate the input and output CM signal of the OTA by
canceling the common-mode signal. This is illustrated in Fig. 3.17(c). The polarity of the
OTA indicated in Fig. 3.16 is only valid in the differential mode (DM) sense. DM
transconductance polarity can be changed by just exchanging the output terminals or
input terminals without adding any extra components. However, the CM
transconductance does not change by exchanging the output or input terminals. In fact,
56
the CM transconductance of the OTA in Fig. 3.16 is always negative. Thus, a loop can be
stable in DM but unstable in CM. Note also that the CMD circuit has to be an inverting
type when used in CMFF to cancel the CM signal at the OTA output (see Fig. 3.17(b)),
and non-inverting when used in CMFB to have negative CMFB loop (Fig. 3.17(c)). The
CMD circuit is illustrated in Fig. 3.18(b). All NMOS (and PMOS) transistors are
matched. The inverting output of the CMD is inverted using the auxiliary circuit to
generate the non-inverting output. If the CMD is used only for CMFF, the auxiliary
circuit is eliminated.
Fig. 3.19(a) shows the I branch of one of the filter biquads. OTA4 provides the LP
output current for the next current-mode filter stage. OTA5 and OTA6 play the same role
as RIF in Fig. 3.11(c) or ωcC in Fig. 3.14. OTA1 and OTA2 form a negative feedback DM
loop, but a positive feedback CM loop. The output node of OTA2, node 1 is a low
impedance (1/Qgm) node due to the resistive connected OTA3. Hence, no CMFB is
needed at this node and only CMFF is used in all the OTA’s that feed this node,
excluding OTA3. If CMFF is used in OTA3, the CM impedance at node 1 will be very
high. Instead, the bias voltage of OTA3 is connected to a fixed voltage, independent of
input and output CM voltages. The use of CMFF in OTA2 breaks the CM loop formed by
OTA1 and OTA2. Without CMFB, node 2 is a CM high impedance node, and hence,
needs CMFB to stabilize it. CMFB loop is formed in OTA1 through CMD2. CMFF is also
used in OTA4, OTA5, and OTA6 to isolate the CM signals in this biquad stage from the
next biquad and from the corresponding biquad in the Q branch. Only two common mode
detectors are needed to form the CM control circuit in this biquad stage with six OTAs.
By using the minimum number of CM control circuits, this efficient scheme saves
57
considerable power and silicon area, and contributes less noise than using a conventional
CM control circuit for each OTA, as in Nauta’s transconductor [26]. The proposed CM
control scheme roughly consumes only about 1/3 of the area and power of CM control
circuitry in conventional schemes. A CMRR in excess of 50 dB is obtained. Fig. 3.19(b)
shows a block diagram for the complex biquad, which consists of two LP biquads, and
two cross-coupled OTA’s for each on the internal nodes 1 and 2.
gm
vi+
vi-
RL
RL
gm
OTA2
OTA1
CMDVCM
NFB
gm
vi+
vi-
gm
OTA2
OTA1
VCM
NFF
CMD
vo+
vo-
(a) (b)
Fig. 3.20 (a) Circuit setup for HD3 analysis (b) Another scenario for HD3 analysis
3.4.2.4. Harmonic Distortion
Due to the long channel used in the OTA, square law characteristics can be assumed
for the MOS transistors. Therefore, it can be shown that while CMFF does not introduce
3rd order harmonics, the CMFB does. To show how CMFB introduces 3rd order
harmonics, consider the simplified case shown in Fig. 3.20(a). OTA1 is loaded with a
linear resistor RL to study the effect of CMFB non-linearity by itself. The CMFB on
OTA1 generates a CM 2nd order harmonic at the output of OTA1. This harmonic mixes
with the DM fundamental output of OTA1 due to the 2nd order non-linearity of the input
58
transistors in OTA2. As a result, 3rd order harmonics appear at the output of OTA2, and
the harmonic distortion can be expressed as:
2
32)1(3
−
−=
TnC
PLmLm
VVVRgRgHD (3.9)
Where VP is the peak voltage of the input signal. Two remarks are drawn in the
above expression: first, the HD3 is inversely proportional to the squared overdrive
voltage. Second, the HD3 due to CMFB vanishes when RL = 0 or RL = 1/gm. However, in
these two cases, CMFB is actually not needed. For gm RL >> 1, the above expression
reduces to:
2
1
2
321
3213
−
=
−
≈TnC
Pout
TnC
PLm
VVV
VVVRgHD (3.10)
Where VPout1 is the signal peak at the output of OTA1. Another practical scenario that
may generate HD3 is shown in Fig. 3.20(b). In this case, OTA1 is loaded with the
resistive connected OTA2.CMFF is used for OTA1, while OTA2 has no CM control. The
HD3 at the output is expressed as:
2
122
21
81/
813
−
=
−
=TnC
PoutP
TnC
mm
VVVV
VVggHD (3.11)
Where VP and VPout1 are the peaks of the signals at the input and output of OTA1,
respectively. Note that in the later scenario, for the same biasing conditions, HD3 is 12dB
worse than the 1st scenario.
3.4.2.5. Filter Architecture
Fig. 3.21 shows the block diagram of the entire complex filter. Passive input high
pass RC filters are used to isolate the CM mixer output from the filter CM input. The
59
voltage VT1 applied through the R of the high pass filter tunes the transconductance of the
filter input stage, which uses the same OTA architecture shown in Fig. 3.16. An
important design issue is how to distribute the gain among the filter stages. If all the gain
(15 dB) is used at the filter input stage, the noise performance will be optimized but the
linearity is degraded and vice versa if the gain stage is placed at the end of the filter
stages. Due to the tough noise requirement on the filter, a 15 dB gain stage is placed at
the filter input as shown in Fig. 3.21. Since Bluetooth uses a frequency modulation
format (GFSK), in-band linearity is not a major issue. In contrast, the filter design should
be focused to improve the out-of-band linearity. Since the out-of-band blockers will be
attenuated by the filter, harmonics generated by the out-of-band blockers are dominated
by the filter’s first gain stage. Hence, to improve the overall filter linearity, the gain stage
is designed to have better linearity than the filter by using larger overdrive voltage (VGS –
VTn) of the input NMOS transistors M1 and M2 shown in Fig. 3.16.
Biquad 1Q=0.52
Biquad 2Q=0.71
Biquad 3Q=1.93
GainStage
+_ +_
+_ +_
vIo+vIo-vQo+vQo-
vIi+
vIi-
vQi+vQi-
VT1
VT1
VT1 Fig. 3.21 The complete 12th order complex filter
60
3.4.3. Frequency Tuning Scheme
3.4.3.1. System Architecture
Fig. 3.22 shows the frequency tuning circuit of the complex filter, which is built to
compensate for process variations. It consists of a relaxation oscillator, two counters to
measure the oscillator and reference frequencies, a comparator, an up-down counter, and
a simple D/A converter (DAC). The relaxation oscillator, discussed in the next
subsection, is based on the same OTA architecture used in the filter. Under nominal
conditions, the frequency of the relaxation oscillator is equal to the reference frequency
(1MHz). The operation of the tuning circuit is described as follows: after system reset,
the 7-b reference and oscillator counters start counting until the reference counter reaches
64. At this time, the Up/Dn counter is clocked to count up or down, or freezes according
to the output of the digital comparator, which compares the content of the oscillator
counter with Dref = 64. The content of the 7-b Up/Dn counter is then converted to an
analog voltage Vc (via a 7 bit DAC) to control the frequency of the oscillator (by
controlling the value of gm though Vb as shown in Fig. 3.16). When the reference counter
overflows (reaches 128), it sends a reset signal to the oscillator counter to begin a new
frequency comparison cycle based on the updated oscillator frequency. Eventually, the
oscillator frequency will reach the reference frequency (the reference frequency is 1MHz
and is derived from the 16MHz crystal oscillator used for the receiver chip) within an
error depending on the DAC resolution. The same control voltage Vc is applied to VCM in
the filter biasing circuits and CM control circuits (Fig. 3.18) to tune the frequency to the
correct value. A dead zone, depicted in Fig. 3.22, is added to the comparator transfer
characteristic to avoid oscillation in the loop around the desired frequency. The width of
61
the dead zone is 3 counter steps around the middle count. The maximum error in the
frequency-tuning loop depends on DAC quantization and the relaxation oscillator
conversion gain. For ±30% process variations and a 7-bit DAC, the maximum frequency
error is ±0.23%. This error is mapped to only 4.6 kHz error in the filter center frequency,
which is quite tolerable for Bluetooth application. The 7-bit DAC is implemented using
resistive string to insure monotonicity and hence stability of the tuning loop. A non-
systematic error should also be considered due to the mismatches between the
transconductance and capacitance in the passive RC LPF and the oscillator. These
mismatches can add roughly 1% error to the frequency tuning. This is equivalent to
another 20 kHz error in the center frequency of the filter, which is still within the
tolerable range of a Bluetooth filter. The advantage of such tuning circuit architecture
over the conventional PLL-based frequency tuning is that it does not need a low-
frequency LP loop filter, which consumes considerable area of the PLL. In addition, it
uses a square wave relaxation oscillator, which is easier to build and to guarantee
oscillations than the sinusoidal oscillator needed in the conventional PLL.
RelaxationOscillator Comparator
=
Up/DnCounter
D/ATo complex
filter
Vc
Systemreset
Dref
Overflow
=64
OscillatorCounter
ClkR
ReferenceCounter
Clk
R OF
Clk
><
Up/Dn
Freeze
freff/16
16MHz
62 63 64 65 66
Comparatoroutput
Oscillatorcounter
Up
Dn
Freeze
7-bit
7-bit
7-bit
Vosc
Fig. 3.22 Frequency tuning circuit for complex filter
62
Guard rings are used to isolate the “noisy” tuning circuit from the filter. An RC LPF
is used at the output of the tuning circuit to further attenuate the noise. A 1 MHz tone was
observed at the output of the filter at 20 dB below the filter input-referred integrated
noise.
3.4.3.2. Relaxation Oscillator
The relaxation oscillator, shown in Fig. 3.23, consists of an OTA, a current switch
(M1-M6), an integrating capacitor, and a fully differential comparator with hysteresis. The
transconductance of the OTA is controlled by changing its common-mode input level
(VC). By applying a constant differential voltage ∆V to the OTA, the output single-ended
current will be given by:
VVVL
WKVgi TnCnmo ∆−=∆= )(' (3.12)
This output current is mirrored to the tail current source of a differential pair. The
output current of the differential pair is integrated on the capacitor CT. The polarity of
that current is controlled by the differential pair transistors. The corresponding slope of
the triangular signal is )4/( Tm CVg ∆ . The capacitor voltage is then compared with VB1 or
VB2 depending on the comparator output. The voltages VB1, VB2, and ∆V are obtained
from the same resistive string of the DAC used to convert the up-down counter content to
analog voltage. The comparator output controls both the differential pair transistors and
the threshold voltage of the comparator itself. The oscillation frequency can be expressed
as:
BT
mosc V
VCgf
∆∆
=41 (3.13)
63
where 21 BBB VVV −=∆ . Since the ratio ∆V/∆VB is determined by ratio of resistors in
the resistive string DAC, it is independent of temperature and process variations and can
be predetermined with good accuracy (depending on matching the DAC resistors).
Hence, the oscillation frequency is proportional to gm/CT with a well-controlled constant
of proportionality. The value of CT is chosen such that, under nominal conditions, the
oscillator runs at the reference frequency when the common mode voltage is at nominal
(1.65V). The feedback tuning loop ensures that the value of gm/C remains constant in the
presence of temperature and process variations.
_
+
_
+
VB1
VB2
M3 M4
M2M1
CT
M6
VDD
vi+
vi-
io+
io-gm
+
+
VC
∆V
−∆V
M5Vosc
io
io
1 : 1
2 : 1
Vdd
0
VB1
VB2
14
m
T
g VslopeC∆
=
4 T B
m
C VT
g V∆
=∆
Fig. 3.23 The relaxation oscillator
3.4.4. Experimental Results
The filter and the frequency tuning circuit have been implemented in TSMC 0.35µm
CMOS process. The chip micrograph is shown in Fig. 3.24. The areas occupied by the
filter and the tuning circuit are 1.6×0.8mm2 and 1×0.4mm2, respectively. The filter
64
operates from 2.7V power supply and draws 4.7mA while the tuning circuit draws
0.8mA. To test the filter frequency characteristics, quadrature sinusoidal signals are used.
Polyphase RC network can be used to generate the quadrature signals [18]. However,
polyphase RC filters can only generate balanced quadrature signals for a narrow
frequency band, and hence, cannot be used to measure the attenuation of interference
signals at positive and negative frequencies (relative to the LO frequency). Furthermore,
process variations may alter the RC time constant, which will lead to unbalanced
quadrature signals. In this case, Tektronix AFG320 signal generator is used to generate
the required quadrature signals as shown in the test setup depicted in Fig. 3.25. Fig. 3.26
shows the filter frequency response for the signal and image sides. The figure shows that
the image rejection ratio is more than 45dB, which is enough for Bluetooth specifications
[1]. The filter attenuates the first and second adjacent channels by 27 and 58dB,
respectively. The filter linearity is quantified in terms of spurious-free dynamic range
(SFDR). The SFDR is measured by applying two tones at the following frequencies:
MHznff C 11 ×+= and MHznff C 22 ×+= (3.14)
Where n is the two tones separation in MHz, Cf is the filter center frequency, 1f is
the frequency of the first tone, and 2f is the frequency of the second tone.
Fig. 3.27 shows the measured SFDR versus n. The in-band SFDR (n=0) is about
45.2dB. Fig. 3.28 shows that the in-band two-tone test from which the in-band SFDR is
measured. Since the filter is followed by a hard limiter (Fig. 3.7), the in-band SFDR is
not a critical parameter in this case. The out-of-band SFDR is a more important
parameter to measure. In Bluetooth, the IP3 is calculated for two interferers at 3 and
6MHz away from the desired signal on one side. The IP3 can be approximately
65
calculated to be about 1.5×SFDR = 91.65dB above the noise floor. The total input
referred noise is 29µVrms and the pass band gain is 15dB. Fig. 3.29 shows the measured
group delay of the filter, from which it is seen that the in-band group delay variation is
about 0.6µs. The asymmetry in the magnitude and group delay responses in Figs. 17 and
20 is a result of parasitic components and mismatches between filter components. Table
3.1 summarizes the experimental results of the filter.
Fig. 3.24 The die photo (filter area = 1.6×0.8mm2 and tuning circuit area = 1×0.4mm2)
ComplexFilter
VIi
VOCM VIi+
VIi-
VQi+
VQi-VQi
VIo+
VIo-
VIo
VQo+
VQo-
VQo
Oscilliscope
OPA3692
OPA3692
ch1 ch2
VOCM
Signal generatorAFG320
ch1 ch2
Amp.
Freq.
Fig. 3.25 Test setup for the complex filter
66
Fig. 3.26 Frequency response at signal and image sides (vertical axis 12dB/div, … ideal,
actual)
Fig. 3.27 The measured SFDR versus n
67
Fig. 3.28 IM3 test for f1 = 1.95MHz and f2 = 2.05MHz
Fig. 3.29 Group delay response
68
Table 3.1 Summarized filter testing results
Parameter Value
Center frequency 2MHz -1dB bandwidth 1.47MHz – 2.53MHz
Pass-band gain 15dB
Input referred noise 29µVrms Vdd 2.7v
Filter current drain 4.7mA
Image rejection ratio >45dB
Attenuation @ MHzfc 1± 29dBc
Attenuation @ MHzfc 2± 58dBc
CMRR > 50dB
In-band SFDR 45.2dB
SFDR at 3 & 6MHz 61.1dB
In-band group delay variation 0.6µs Area (filter + tuning circuit) (1.28 + 0.4)mm2
3.5. Conclusions
A pseudo differential OTA-C complex filter design for low-IF Bluetooth receiver has
been presented. The main highlights for the design are: (i) a pseudo differential OTA is
used to comply with low voltage operation, (ii) a sound scheme for common-mode
control is implemented using a minimum number of CMFF and CMFB circuits.
Compared to using a separate CM control circuit for each OTA [20], the proposed CM
control scheme roughly consumes only about 1/3 of the area and power of CM control
circuitry in conventional schemes, (iii) A non-conventional frequency tuning circuit
architecture is used that has advantages over the conventional PLL, in terms of silicon
area and design complexity.
69
CHAPTER IV
DIRECT DIGITAL FREQUENCY SYNTHESIZER (DDFS)
In modern wireless communication systems, fast frequency switching with fine
frequency steps is crucial. An example of such a system is Bluetooth, where the signal
modulation is GFSK with about 160kHz frequency deviation. A simplified diagram of
the Bluetooth transmitter is illustrated in Fig. 4.1, where the input bit stream is pulse
shaped by the Gaussian filter and then frequency modulated with frequency deviation (fd)
of 160kHz. The frequency modulator has to provide quadrature outputs that are needed
for the quadrature mixer operation.
Date ind(t)
X
X
90o
GaussianFilter
FrequencyModulator Σ PA
g(t)
cos(fdg(t)t)
sin(fdg(t)t)
Quadraturemixer
sin(ωLOt)
cos(ωLOt)cos((ωLO+fdg(t))t)
Fig. 4.1 Bluetooth transmitter architecture
There are several possible implementations for the frequency modulator. A
traditional PLL based synthesizer is not suitable in these applications due to the inherent
loop delay. Another limitation of a PLL is the small range of frequency locking and the
70
limited frequency resolution. Open loop voltage controlled oscillator (VCO) is also not
suitable due to the limited control on the output frequency. A direct Digital Frequency
Synthesizer (DDFS) is the most suitable implementation due to its fast frequency
switching, with small frequency steps over a wide band. Note also that PLL and VCO
implementations are not suitable for the frequency modulator in Fig. 4.1 since the output
frequency centered around 0. Since the DDFS implements the sine and cosine functions
directly, it can be used in frequency modulator in Fig. 4.1. Finally, the DDFS can
generate quadrature output phase very accurately, unlike the PLL where mismatches
between components may result in phase mismatch between quadrature outputs. The
implementation of the quadrature output frequency modulator is described in detail in the
next few sections.
4.1. Conventional DDFS Architecture
Conventional ROM-based direct digital frequency synthesizers (DDFS), as shown in
Fig. 4.2, are able to meet the above requirements by storing the values of the sine
function in a ROM and scanning these values at a rate proportional to the desired
frequency. The digital ROM output is converted to analog using a digital to analog
converter (DAC).
The main factors that determine the signal purity in this architecture are: (1) the
phase quantization due to finite resolution of the phase accumulator, (2) amplitude
quantization noise due to finite resolution of the DAC, and (3) static and dynamic non-
idealities of the DAC. The ROM size is exponentially proportional to the desired phase
resolution, resulting in huge area consumed by the ROM for reasonable phase
resolutions. Moreover, the ROM should be addressed at a much higher rate than the
71
desired output frequency for moderate spectral purities. Scanning the ROM at high speed
makes it power hungry and then, unsuitable for portable wireless applications.
RO
M
Com
plementor
N
LinearDACN
OuputSine
LPF
Adder
Latch
Register
L
L
L
PhaseAccumulator
fclk
D
L M-2
Com
plementor
2nd MSB
M M-2
MSB
2MSBs
Fig. 4.2 Conventional ROM-based DDFS
Several attempts have been made to reduce the ROM size by using various
techniques. The first category of solutions is based on trigonometric identities; the
simplest of which is the quarter wave symmetry in the sine function (see Fig. 4.2). Other
trigonometric formulas have been used to split a large ROM into two smaller coarse and
fine ROMs [27]. The second category of solutions approximates the sine function over
the first quarter period by another function, f(x), that can be easily implemented, as
illustrated in Fig. 4.3. In this implementation, a ROM look-up table is used to store the
error, sin(x)-f(x), which results in less memory word-length requirements. The simplest
form is the sine-phase difference method [28] that uses a straight-line approximation for
the sine function, i.e., f(x)=x. In this scheme 2 bits of memory word-length are saved.
Parabolic approximation has been also introduced [29], which results in saving 4 bits of
memory word-length.
However, the above techniques are still consuming considerable area and power,
hence not suitable for low cost portable applications. The third category of solutions is to
72
use a combination of small ROM to store few sample points and a linear interpolation
between these points for full computation of the generated sine function as shown in Fig.
4.4. It has been shown that this technique is efficient and the hardware cost required for
the additional calculations is shown to be lower than the first two categories [30].
f(x)x
ErrorROM
sin(x)-f(x)
Adder
LinearDA
C
OuputSineLPFA
dder
Latch
Register
L
L
L
PhaseAccumulator
fclk
D
L M
Fig. 4.3 Compression technique using approximation f(x) of the sine function
Another approach was adopted in [31] to avoid using ROM by using nonlinear DAC
as shown in Fig 4.5. In this architecture, the non-linear DAC is used to achieve the
function of the phase to amplitude conversion and the digital to analog conversion at the
same time. This approach was shown to provide considerable area and power savings
compared to the conventional approach since the ROM is removed. There were two
options for the DAC implementation.
The first implementation is based on resistive string, which consumes less power but
is inherently slow (fclk=25MHz) and occupies significant area (1.7mm×1.7mm in 0.5µm
technology) due to the large number of resistors and transistors used. The second
implementation of the nonlinear DAC is using current-mode techniques to enhance the
speed at the expense of power consumption (92mW at fclk=230MHz). To further reduce
73
the power consumption and die area, a technique was proposed in [32] to split the
nonlinear DAC into a coarse DAC and a fine DAC.
In this chapter, a different approach is chosen to implement ROM-less DDFS based
on piecewise linear approximation of the sine function as will be presented in the next
section [33]. The proposed architecture is shown to have significant area and power
savings at high clock rates. Design considerations of the building blocks will be discussed
in sections 4.3-4.5. Testing results of the DDFS, as well as a comparison with recently
published work, will be presented in section 4.6. Finally concluding remarks are drawn in
section 4.7.
Fig. 4.4 Combining a small ROM and linear interpolation to generate sine function
M-2
Com
plementor
Non-L
inearD
AC
M-2
2nd MSB
MSB
OuputSine
Adder
Latch
Register
LL
L
L
PhaseAccumulator
fclk
M
D
Fig. 4.5 ROM-Less DDFS using non-linear DAC
74
4.2. Proposed DDFS Architecture
The proposed architecture is based on the idea of breaking the sine function into
linear segments as shown in Fig. 4.6, where four segments are shown for the purpose of
illustration.
1
0 x0 x1 x2 x3
k0
k2
k3
k1
ideal
approximated
sin(x)
P(x)
Fig. 4.6 Sine wave approximation
For a given number of segments, the segments' slopes (k0, k1, ……) are chosen to
minimize the integrated mean square error between the ideal, sin(x), and the approximate
piece-wise linear, P(x), curves. In order to simplify the implementation of this
approximation, the number of segments is chosen to be in powers of 2. The points Xi (i =
0,1, ..…) are selected to be equally spaced to further simplify the design. A MATLAB
code (Appendix A) is developed to determine the optimum set of slope values, i.e., given
the number of piecewise segments, the slope values yielding minimum mean square error
(MMSE) are determined. Where MMSE is expressed as:
75
[ ]∫=
−=2
0
2)()sin(min
πx
dxxPxMMSE (4.1)
The piecewise linear function is implemented using the block diagram shown in Fig.
4.7. The most significant M bits of the phase accumulator output are used in the phase-to-
amplitude (P/A) converter. The P/A converter consists of the complementor, the linear
DAC, and the switched weighted-sum (SWS) blocks. The first two MSBs of the
accumulator are used to select the quarter in the sine wave cycle. The next M-2 bits are
fed into the complementor, whose output is then split into two parts: α and β where α is
the MSB part (a bits long), and is corresponding to the segment number, and β is the LSB
part (b bits long) which is applied to the input to the DAC. For a given phase resolution
(M), the proposed architecture uses an (M-2-a) bits linear DAC. Whereas, in the
architecture presented in [31], an (M-2) bits nonlinear DAC is needed. Reducing the
DAC bits saves significant area and promotes faster operation. The output of the SWS
block is given by:
0
1
min
12
1max
1
0min
==
=++= ∑∑−
+=
−
=
EforVV
EforVkVkVkV
sws
nn
nDACnsws
a
α
α
α (4.2)
Where VDAC is the linear DAC output in the range of Vmin to Vmax , and E is a digital
signal that enables the SWS block depending on which half cycle of the sine function,
positive or negative, is currently running. The SWS block in each branch implements half
of the sine wave cycle, and hence the differential output voltage is a full sine wave. The
frequency of the output sine wave is given by:
min.2.
fDfD
f Lclk
out == (4.3)
76
Where D is the digital accumulator input, L is the number of bits in the accumulator,
fclk is the input clock frequency, and fmin is the minimum synthesized frequency
(frequency resolution) that can be obtained.
Adder
Latch Register
Com
plementor
Linear
DA
C
SwitchedWeighted-
Sum
Vmax
Vmin
VDACL M-2
2MSBs
b
a
Vsws
L
L
L
PhaseAccumulator
D
M
2nd MSB
MSB
+
Vsws
_
a MSB
P/A Converter
E
Efclk
Vsin=Vsws-Vsws+ _
M=a+b+2
α
α
β
SwitchedWeighted-
Sum
Fig. 4.7 Block diagram of the proposed DDFS architecture
In the proposed design, L=16 and fclk=100MHz yielding a frequency resolution of
1.5kHz. To allow a high clock rate, a 4-stages carry look-ahead adder (4 bits/stage) is
used in the accumulator. Sources of distortion in the above architecture are the limited
number of segments (2a), phase resolution (M), and slope resolution. The effect of each
of these parameters on the performance is simulated using MATLAB. Fig. 4.8 shows the
effect of the number of segments on the spurious free dynamic range (SFDR). Note that
the SFDR improves by 12 dB by doubling the number of segments. The SFDR is 59dBc
and 71dBc for 8 (a=3) and 16 (a=4) segments, respectively.
In Bluetooth, the transmitter spurious emissions should be less than –20dBc to -
40dBc at 2MHz and -40dBc to -60dBc at 3MHz, depending on transmitter power class.
Since the 16-segments DDFS consumes about 50% more power and area than the 8-
segments design, as will be shown at the end of section 4, 8-segments design is adopted.
77
A passive RC pole can be used at the DDFS output to attenuate the spurs at frequencies
far from the fundamental output. For 8 segments, the effect of the finite phase resolution
is shown in Fig. 4.9. It shows that no significant improvement in the SFDR is achieved
for phase resolutions more than 10 bits. For this choice of M=10 and a=3, only 5-bit
DAC (b=M-2-a=5) is needed.
Fig. 4.8 Effect of the number of segments on the SFDR
Fig. 4.9 Effect of phase resolution on the SFDR
78
The weighted-sum function is implemented using resistive dividers (as will be
discussed later). Each resistance is an integer multiple of a unit resistance ∆R, which
determines the slope resolution. The effect of finite unit resistance is shown in Fig. 4.10
for 8 segments and 10 bits of phase resolution.
Note that for a normalized unit resistance (∆R/Rmax) less than 0.4%, there is no much
improvement in SFDR. To allow for some margin, (∆R/Rmax)=0.2% is a good choice for
an 8 segment sine shape. An extra care must be taken in the layout of these resistors to
achieve the required resolution.
Fig. 4.10 Effect of finite unit resistance on the SFDR
Table 4.1 summarizes the required values of phase resolution (M), normalized unit
resistance, and the corresponding SFDR for different number of segments. Note that
doubling the number of segments, i.e., increasing ‘a’ by 1 requires 2-bit increase in the
phase resolution. This implies doubling the linear DAC size, i.e., increasing ‘b’ by 1.
79
Quadrature outputs are generated by replicating the P/A converter. The 900 phase shift is
implemented by adding 01 to the two MSBs of the accumulator output.
Table 4.1 Required values of (M), (∆R/Rmax), and corresponding SFDR for different number of segments
Number of segments
Phase resolution (M)
(∆R/Rmax) SFDR
4 8 bits 0.7 % 47 dBc 8 10 bits 0.4 % 59 dBc
16 12 bits 0.2 % 71 dBc 32 14 bits 0.1 % 83 dBc
4.3. Linear DAC
The 5-bit resistive string DAC is implemented as shown in Fig. 4.11. In order to
have smoother transitions at the corner points between segments in the output sine wave,
a ½ LSB offset is introduced to the DAC output by using R/2 value for the lower and
uppermost resistors. A 5-to-32 decoder is used to turn ON the switch corresponding to
the digital input β. At any given time, only one switch must be turned ON. Due to non-
equal delays at the decoder outputs, more than one switch can be turned ON at the same
time or all the switches can be turned OFF. This will result in undesirable glitches at the
output of the DAC. To solve this problem, the outputs of the decoder are sampled at
CLK .
The settling time for each digital input word is determined by three factors: (1) The
total capacitance (CL) at the output node VDAC which is dominated by the drain
capacitances of all the switches and the input capacitance of the next stage. (2) The input
resistance of the resistive string seen from the corresponding node (m). The closer the
node to the middle of the resistor string, the higher the resistance seen. The value of R is
80
chosen as a trade-off between power consumption and settling time. For this particular
design, R=50Ω is found to be a good compromise. (3) The ON resistance, RON, of the
corresponding switch Mm. In order to minimize the load capacitance CL, different
transistors’ widths are used to have the same settling time for all digital input
combinations. To allow for large overdrive voltage in the NMOS switches, the terminal
voltages (Vmin and Vmax) of the resistor string are chosen to be 0V and 0.5V, respectively.
R/2
R
R
R
R
R
R/2
R
Y0
Y1
Y2
Y29
Y30
Y31
VDAC5 BitDecoder
Y31
b0
b1
b2
b3
b4
b
Vmax=0.5V
Vmin=0V
R=50
mMm
Ym
CL
Ω
Ym
Y0
CLK
Fig. 4.11 Linear DAC circuit implementation
81
4.4. Switched Weighted-Sum Block
Two SWS blocks are used in the system, one for each half of the sine wave cycle.
The details of the SWS block are shown in Fig. 4.12.
Analog
DeM
UX
WeightedSum
Vmax
Vmin
VDAC
a
Vsws
V0
V2a-1E
α
Analog Demultiplexer Cell i=0,2a-1
VmaxVDACVmin
ViControlLogic
a
E
Enable
Vn
Fig. 4.12 Switched weighted-sum block
The output of the SWS block can be written as:
∑−
=
=12
0
a
nnnsws VkV (4.4)
The role of the analog demultiplexer is to route Vmax, VDAC, or Vmin to Vn based on
the value of α, corresponding to the segment number, and enable E as follows:
0 OR)1&(
1&1&
min
max
==>=====<=
EEnforVVEnforVVEnforVV
n
DACn
n
ααα
(4.5)
The analog demultiplexer consists of 2a cells, one for each output. The basic analog
demultiplexer cell for each output Vi for i=0, 2a-1 is illustrated in the same figure.
The weighted-sum function described in equation (4.4) is implemented using
resistors and buffers as shown in Fig. 4.13. Assuming ideal buffers (unity gain and zero
output resistance), the weighted-sum output Vsws is given by:
82
( )
( )
( )
T
nnn
nn
nnn
sws G
RV
R
RVV
a
a
a
∑
∑
∑−
=
−
=
−
= ==
12
012
0
12
0
/
/1
/ (4.6)
Where RT=(1/GT) is the parallel combination of all resistors in the weighted-sum
network. Comparing equations (4.4) and (4.6), we get:
12,,1,0 −== a
T
nn n
RRk L (4.7)
Equation (4.7) is used to obtain the values of Rn from the segments’ slopes. Note
from equation (4.6) that Vsws depends on the ratio of resistors, hence the matching of the
resistors in the weighted-sum block in each branch is a critical issue. On the other hand,
matching between the resistors of the weighted-sum blocks in both positive and negative
branches is not critical.
The buffers are the most power consuming parts in the proposed system. Care must
be taken in the design of these buffers to minimize the overall power consumption while
keeping good linearity.
Fig. 4.13 shows also the transistor level design of each buffer. PMOS input
transistors are used to eliminate the body effect by shorting the source and body terminals
(nwell technology is used). Another advantage of using PMOS input transistor is that the
input voltage can be as low as 0V. Since this voltage is routed from the DAC resistor
string to the buffer input using MOS switches (in the DAC and the analog demultiplexer),
then high-speed NMOS switches can be used with high overdrive voltage. This helps to
minimize the delay associated with these switches without having to use wide transistors.
Note that the output voltage of each buffer is DC shifted by VSG relative to its input.
Since all the buffers are biased using the same biasing voltages VB1 and VB2, they all
83
have the same nominal DC shift. However, due to mismatches between transistors in
different buffers, this DC shift may differ from one buffer to the other. These DC shift
mismatches will result in a slight offset in the output sine wave, but will not affect its
spectral purity.
Since buffers have different load resistances, the actual relative slope values of the
segmented sine wave will be slightly different from equation (4.7) if the buffers were
identical due to the finite buffer output resistance. One way to dilute this effect is to
reduce the buffers output resistance by increasing the bias current or increasing W/L of
M1 (Fig. 4.13), which will increase the power consumption or increase the load
capacitance to the analog demultiplexer, respectively.
R0V0
R1V1
Vws
V2a- 1
R2a-1
Buffer
Buffer
Buffer
Vn Vws
VDD
VSS
VB 1
VB 2
Vi n
Vout
M1
M2
M3
Buffer sub-cell
Fig. 4.13 Implementation of the weighted-sum block
Instead, to have the same loading effect on all buffers, each buffer is designed such
that the ratio of its output resistance to the load resistance is the same for all buffers. To
84
allow better matching between buffers, each buffer is designed as a number of parallel
buffer sub-cells. These sub-cells are identical in all buffers, but the number of parallel
sub-cells in each buffer is inversely proportional to the desired output resistance. As a
result, the buffers' output resistances will account for a slight attenuation, but the relative
slope values of the segments remain unchanged. Consequently, the total current drain in
all buffers will be inversely proportional to the parallel combination, RT, of all resistors in
the weighted-sum network. Hence, scaling of all resistors is very important in
determining the overall power consumption. However, the higher the value of RT, the
slower the response of the circuit due to the large time constant RTC, where C is the input
capacitance of the following stage. The settling error within the clock period is given by:
CRfout
TclkeVerrorsettling1
−
∆= (4.8)
The largest error corresponds to the largest output voltage step VV 5.0max =∆ . This
largest error should be set to be smaller than the smallest output step, which corresponds
to the smallest slope. For the particular case when a=3, the smallest slope value (obtained
from the MMSE algorithm implemented in MATLAB) is 0.096 for a sine wave of unit
amplitude. Therefore, the smallest output step is:
mVV 5.1096.02
5.0slopesmalleststepDACstepoutputsmallest 5min =×=×=∆= (4.9)
By fixing the settling error in equation (4.8) smaller than the above value, we get:
clk
clk
T fCVV
fCR
.81.51
ln.
1
min
max
=
∆∆
< (4.10)
The value of C determines the integrated output noise due to resistors, known as
KT/C noise. This noise should be sufficiently smaller than the largest output spur, which
85
in this case is 59dB below the fundamental tone. If we set the noise level to be lower than
this spur, then:
)20/(102 SFDR
AC
KT< (4.11)
Where, K is Boltzmann constant, T is the temperature (assumed 300°K), and A is the
output amplitude (0.5V).
In case of 8 segments approximation, the above equation yields C > 0.026pF. This is
rather a loose condition for this capacitance, which is typically larger than 0.1pF. In this
design, 0.3pF is assumed for the load capacitance, for which the maximum parallel
resistance RT is 4.7kΩ, as given by equation (4.10) for fclk=100MHz. To allow for some
margin for process variations, all resistances are scaled to have a nominal parallel
combination of 3kΩ. The lowest and highest resistors in the weighted sum network are
15kΩ and 150kΩ, respectively.
An alternative switched capacitor implementation of the weighted-sum block is
shown in Fig. 4.14. The cycle period of the two phases ϕ1 and ϕ2 is (1/fclk). Since the
phases ϕ1 and ϕ2 are non-overlapping phases, the on-time period should be chosen less
than (0.5/fclk). The weighted-sum output Vsws is given by:
∑−
=
−=
12
0
1 a
nnn
Fsws VC
CV (4.12)
Equation (4.12) is valid for phase ϕ1, while the output is hold by capacitor CH during
ϕ2. Comparing equations (4.4) and (4.12), we get:
12,,1,0 −== a
F
nn n
CC
k L (4.13)
86
Equation (4.13) is used to obtain the values of Cn from the segments’ slopes. Note
from equation (4.13) that Vsws depends on the ratio of capacitors, hence the matching of
the capacitors in the weighted-sum block in each branch is a critical issue. On the other
hand, matching between the resistors of the weighted-sum blocks in both positive and
negative branches is not critical.
_
+
C2a
-1V2
a-1
V0
ϕ 2
ϕ1C0
V1
ϕ 2
ϕ1C1
ϕ 2
ϕ1
ϕ2
ϕ1CF
ϕ 1
ϕ2CH
Vsws
Fig. 4.14 Alternative implementation of the weighted-sum block
Example: The minimum value of the capacitor Cmin is determined by the noise
requirement, which should be sufficiently smaller than the largest output spur. For the
case of 8 segments the integrated noise should be 59dB below the fundamental tone.
According to equation (4.11) this results in a Cmin of 0.3pF. Because of the finite GBW of
the OpAmp, the output voltage will not settle to its final value during ϕ1. To make this
error negligible, it is required, as a rule of thumb, that:
5>⋅⋅ TGBWm (4.14)
87
Where m is the capacitor ratio between the sum of all the feedback capacitors
divided by the sum of all the capacitors connected to the input terminal of the Op Amp,
and T is the period of the clock frequency. In our case, m=1/5 according to equation
(4.13). Thus, from equation (4.14) and for fclk=100MHz, GBW should be larger than
2.5GHz.
GHzCV
ICg
GBWov
tailm 5.22
105.22
105.2min
9
min
9 >⇒×>⇒×>ππ
(4.15)
A value of Vov=0.5V and Cmin=0.3pF results in a tail current of the input stage of the
OpAmp Itail of 2.4mA. Due to the finite gain of the OpAmp, the steady state settling value
deviates from its ideal value. To minimize this error, the gain of the OpAmp should be at
least 60dB, which might require a two-stage OpAmp. This makes the design more
difficult in terms of stability requirements, i.e., phase margin. For the OpAmp to act as a
single pole system with enough phase margin, the frequency of the non-dominant pole
ωnd must be pushed at least twice that of GBW, i.e., ωnd > 5Grad/sec. In the used
technology (0.5µm process), this means even pushing more current than calculated
previously, i.e., Itail. The current consumption of the first solution of the weighted sum
block, shown in Fig. 4.13, is 1.5mA as shown in section 5. Since the current consumption
of the second solution, shown in Fig. 4.14, is more than 2.4mA, the first solution has been
adopted in the final design.
Design procedure: A simple design strategy for the proposed DDFS architecture, for a
given SFDR, can be itemized as:
88
(1) Choose the number of segments of the piecewise linearly approximated sine
function according to Fig. 4.8 for the required SFDR.
(2) The values of the slopes of the piecewise linearly approximated sine function are
calculated to minimize MMSE as given by equation (4.1)
(3) The required values of phase resolution (M) and resistance mismatch (∆R/Rmax)
are calculated from Table 4.1.
(4) Calculate the value of C such that the integrated output noise, i.e. KT/C noise, is
sufficiently smaller than the largest output spur as given by equation (4.11).
(5) Calculate the value of the maximum parallel resistance RT such that the settling
error is smaller than the minimum output step, ∆Vmin, for the used clock
frequency fclk.
Example: Consider the case of 16 segments (a = 4). The targeted SFDR is 71dB and the
smallest slope is 0.05. According to Table 4.1, the minimum phase resolution for 71dB
SFDR is 12bits. Hence, the number of DAC bits is 62 =−−= aMb . From equation
(4.11), we get C > 0.42pF. If we take C = 0.5pF (this may require adding a physical
capacitance), then the parallel combined resistance will be RT < 2.8kΩ. If we choose RT =
2kΩ, this design will consume about 50% more power compared to the 8-segments
design. The buffers’ area will roughly increase by the same percent.
4.5. Experimental Results
The proposed quadrature (I & Q) output DDFS has been implemented through
MOSIS, in 0.5µm AMI CMOS process. The die photo is shown in Fig. 4.15. The chip
active area is 1.4mm2, of which 25% is occupied by the phase accumulator. On-chip
buffers are included to drive the pin capacitance. The DDFS operates from a single 2.7V
89
supply with 3mA current drain and a clock frequency of 100MHz. The testing setup
incorporates an off-chip instrumentation amplifier, as shown in Fig. 4.16, for differential
to single-ended conversion. It uses low distortion (THD=-75dB at 1MHz) buffer
amplifiers with 1pF input capacitance.
Fig. 4.17 shows the single-ended outputs (Vsws+and Vsws
-) of the I branch as well as
the differential output when fout=98kHz. Fig. 4.18 shows the two quadrature outputs of
the I & Q branches at the same output frequency. The peak-to-peak magnitude is about
910mV, which is slightly less than the ideal 1Vp-p magnitude due to the expected
attenuation of the weighted-sum and the output buffers.
The modulation capabilities of the proposed DDFS have been also tested. Fig. 4.19
shows an example of frequency modulation where the modulating signal is a square wave
of frequency 1kHz. An example of amplitude modulation is shown in Fig. 4.20. The
modulating signal is applied at Vmax (see Fig. 4.10). It is a sine wave of frequency 1kHz
and peak-to-peak magnitude of 400mV with a DC offset of 250mV. The output spectrum
for fout=(fclk/1024) = 98kHz and fout = (fclk/64) = 1.56MHz, at which the SFDR is 57.3dBc
and 42.1dBc, are shown in Fig. 4.21 and Fig. 4.22 respectively. The SFDR versus the
output synthesized frequency is plotted in Fig. 4.23. The SFDR is better than 59dBc for
low synthesized frequencies. For high synthesized frequencies, the SFDR is degraded due
to large output steps of the DAC and the switched weighted-sum blocks. Fig. 4.24 shows
the SFDR versus the clock frequency for fout=(fclk/128).
The DDFS is shown to operate from a clock frequency up to 130MHz. Higher clock
frequencies could not be achieved due to the frequency limitations of the phase
accumulator.
90
Fig. 4.15 Chip micrograph
DDFSIC
V+SWS
R2
R2R2
R2
VIo
Oscilloscope
Signal generator
CLK
V-SWS
DIP Switch
Fig. 4.16 Testing setup
91
Fig. 4.17 Single-ended and differential outputs of the I branch (fout=98kHz)
Fig. 4.18 Quadrature outputs I and Q at fout=98kHz
92
Fig. 4.19 Frequency modulation
Fig. 4.20 Amplitude modulation
93
Fig. 4.21 Output spectrum at fclk=100MHz and fout=98kHz with SFDR=57.3dBc
Fig. 4.22 Output spectrum at fclk=100MHz and fout=1.56MHz with SFDR=42.1dBc
94
Fig. 4.23 SFDR versus synthesized frequency at fclk=100MHz
Fig. 4.24 SFDR versus clock frequency at fout= fclk/128
95
The proposed DDFS is compared with the recently reported synthesizers [30-32] as
listed in Table 4.2. Note that the DDFS presented in [30] does not have an On-chip DAC
and the DDFS presented in [32] does not have quadrature outputs. The energy of the
proposed DDFS (in mW/MHz) is significantly lower than state-of-the-art
implementations due to the removal of the ROM and the small DAC size. Therefore, the
proposed design is more suitable for low power portable applications. A better SFDR can
be achieved, using the proposed architecture, by doubling the number of segments (a =
4). This can increase the SFDR by 12dB at the expense of increasing the power
consumption and active area by roughly 50% and 40%, respectively.
Table 4.2 Comparison with recently published work
[30] [31] [32] This Work
Clock frequency (MHz)/ Operating
frequency (kHz)
30
1560
230
200
300
4000
100
200
Phase resolution (bits) 12 10 12 10
Technology (µm CMOS) 0.8 0.5 0.25 0.5
Quadrature outputs Yes Yes No Yes
On-chip DAC No Yes Yes Yes
Power dissipation (mW) 9.5 92.5 240 8
Power supply (V) 3.3 3.3 2.5 2.7
SFDR (dBc) 60 55 62 59
Active area (mm2) 0.9 1.6 1.4 1.4
Energy (mW/MHz) 0.3 0.4 0.8 0.08
96
4.6. Particular Case
The conventional direct digital synthesizer, i.e., ROM based, can been used to
provide sawtooth and ramp waveforms as presented in [34]. The same applies for the
proposed DDFS. This is the case because the proposed DDFS share the same front end as
the conventional DDFS. The front end is composed of a digital part (the phase
accumulator and the complementor) followed by a D/A converter, as shown in Fig. 4.25.
Please note that the frequency of the saw-tooth and triangular waveforms, shown in Fig.
4.25, is 4fout and 2fout, respectively, where fout is defined in equation (4.3). The peak of the
triangular wave after the linear DAC is Vmax, while the slope is 1.
One of the possible applications of using the ramp function is in built-in self testing
(BIST) [35]. The basic requirement in those kinds of applications is the accuracy of the
ramp waveform. This translates directly to the accuracy of the DAC in Fig. 4.25 since the
digital part does not play any role in the distortion of the waveform. Many DAC
architectures have been reported in the literature [36]. For the resistor string approach,
discussed in section 4.3, the accuracy depends on the matching precision of the resistors.
Using polysilicon resistors can result in up to 0.1%, i.e., 10 bits of accuracy. This might
not be enough for some applications. Dynamic techniques with current switching can be
used to realize D/A converters with higher accuracy (up to 16 bits) [36].
97
Adder
Latch
Register
Com
plementor
Linear
DA
C
Vmax
Vmin
VDACL M-2
2 MSBs
bL
L
L
PhaseAccumulator
M
2nd MSB
fclk
β
Fig. 4.25 Front end of DDFS
4.7. Conclusions
A low-power ROM-less quadrature DDFS architecture has been presented. It uses a
piecewise linear approximation of the sine function. The proposed DDFS has been
implemented in 0.5µm CMOS technology and occupies an area of 1.4mm2. A 16-bit
frequency control word results in a tuning resolution of 1.5kHz at a 100MHz clock
frequency.
The proposed design operates from a single 2.7V supply while consuming 8mW.
The design features an SFDR that is better than 50dBc for synthesized frequencies up to
fclk/256. The proposed architecture also incorporates different modulation capabilities.
The modulation formats include frequency modulation and amplitude modulation. Since
the proposed design consumes significantly less power than recently reported designs, it
is a good candidate for wireless portable communication applications that use frequency
modulation such as Bluetooth and GSM.
98
There are two frequency limitations in this design. First, higher clock frequencies are
limited by speed of the phase accumulator. Digital techniques can be used to solve this
problem [31]. Using better process, i.e. smaller feature size, improves the speed of digital
blocks such as the accumulator. Second, higher synthesized frequencies are limited by the
settling time of the DAC due to large output steps. The values of the resistors in the DAC
can be reduced to improve the settling time at the expense of power consumption.
For applications that require higher SFDR, the number of linear segments can be
increased at the expense of area and power consumption. For example, doubling the
number of segments increases the SFDR by 12dB at the expense of increasing the power
consumption and active area by roughly 50% and 40%, respectively. The proposed
architecture can be also used to generate triangular and/or saw-tooth waveforms, as
shown in Fig. 4.7, using an extra DAC and a smoothing low pass filter.
99
CHAPTER V
CHAMELEON: A MULTI-STANDARD RECEIVER DESIGN
The differences between Bluetooth and Wi-Fi standard features (data rate, security,
communication protocol, etc.. ) lead to a natural partitioning of applications. However,
many electronics devices nowadays, such as laptops, pocket PCs, cell phones, digital
cameras, and PDAs, support both Bluetooth and Wi-Fi standards to cover wide range of
applications. The cost of supporting both standards is a major concern. Therefore the
need for a dual-mode transceiver is evident. To keep the size, and therefore the cost, of
such system at a minimum, as many building blocks as possible must be shared in both
operating modes. Since Bluetooth and 802.11b share the same RF frequency band, it may
be possible to come up with a receiver architecture that can be configured to adopt both
standards. RF front end can be shared between both standards. However, since the two
standards use different modulation format and have different channel bandwidths, sharing
of the baseband blocks is not that easy.
5.1. Possible Receiver Architectures
In section 3.3, we found that low-IF receiver architecture is most suitable for the
Bluetooth standard. On the other hand, due to the large channel bandwidth in Wi-Fi, a
direct-conversion receiver (DCR) is the best approach. Therefore, building the receiver
that supports both standards is not as straightforward. Fundamentally, there are three
possible combinations of architecture of both standards when implementing the multi-
standard Bluetooth/Wi-Fi receiver:
100
(1) DCR for Bluetooth and Wi-Fi
(2) Low-IF for Bluetooth and DCR for Wi-Fi
(3) Low-IF for Bluetooth and Wi-Fi
The first combination is the best in terms of area and power consumption. The area is
minimal because both standards use the same architecture, which means maximum level
of building blocks sharing. However, problems associated with using the DCR
architecture for Bluetooth (as discussed in section 3.1) must be solved. In the second
combination, each standard uses its best architecture which means easier implementation
for each standard at the expense of increased cost because of the reduced sharing of
building blocks between both standards. This approach has been used in [37, 38]. The
third combination uses the low-IF architecture for both standards. Due to the large Wi-Fi
channel bandwidth, IF frequency has to be large enough resulting in high power
consumption, especially in Wi-Fi mode [39].
For combination (2), a couple of these solutions are found in literature [37, 38]. The
solution [37] uses DCR architecture for Wi-Fi and low-IF architecture for Bluetooth as
shown in Fig. 5.1. The RF part (LNA, mixer, and LO) is shared between the two
standards while separate baseband parts (Filter, amplifier, and demodulator) are used. An
analog multiplexer (SEL) is needed to switch the mixer output either to the Bluetooth or
Wi-Fi baseband circuitry. As mentioned before, this approach is not optimum in terms of
silicon area. Separate baseband circuitry is used to claim simultaneous operation.
However, simultaneous operation is practically not possible using the same RF front-end.
This is simply because there might be a situation when there is a strong signal in one
standard, requiring low gain in the front-end, and weak signal in the other, requiring high
101
front-end gain. Furthermore, using the same LO is not acceptable because it is not
guaranteed that Bluetooth and Wi-Fi will have the same RF frequency.
Fractional-NSynthesizer
BT/802.11b
SEL
Demod
802.11b
BT
ClockGenerator
I
Q
I
Q
LNA
AnalogMultiplexer
Fig. 5.1 Dual mode receiver architecture used in [37]
The second solution [38] uses double down-conversion architecture as shown in Fig.
5.2. This architecture is used to avoid the LO self-mixing dc offset problem in the
receiver and the injection-locking problem between the PA and VCO in the transmitter.
The signal at the LNA output is mixed down to 800MHz by the first LO 1.6GHz and then
mixed down again to baseband with 800MHz quadrature LOs. In Bluetooth mode, the
signal is down-converted to a 2MHz low-IF for channel-select band-pass filtering and
limiting. In Wi-Fi mode, the signal is down-converted to dc for channel-select low-pass
filtering and signal amplitude control with VGA’s. A dual-mode channel-select filter is
used to save silicon area. A 5th order Butterworth complex filter with 1MHz bandwidth is
used in Bluetooth mode. In Wi-Fi mode, a real filter reconfigured as a fifth order
Butterworth LPF with 7.5MHz bandwidth by frequency scaling the filter poles and
eliminating the cross-coupling between I and Q branches (see section 3.4.1).
Furthermore, the baseband amplifier stage is configured as a four-stage limiter with ac
102
coupling in Bluetooth mode. In Wi-Fi mode the amplifier stage is configured as a four-
stage VGA. Theoretically, due to baseband circuit reuse in both modes, the receiver in
[38] should occupy less area than the receiver in [37]. However, since the two modes
have different receiver architectures, there is still some area overhead that is needed to
reconfigure the baseband blocks for both standards.
802.11b
LO2-I
LO2-Q
BTBT
802.11b
I
Q
From RF (802.11b-RX)I-Out
(802.11b-RX)Q-Out
(BT-RX)I-Out
(BT-RX)Q-Out
Fig. 5.2 Dual mode receiver architecture used in [38]
5.2. Proposed Direct-Conversion Receiver Architecture
As discussed in the previous section, using the same receiver architecture for both
standards allows maximum level of block reuse to save silicon area, which is the main
objective of having a dual-mode receiver. The proposed system architecture is shown in
Fig. 5.3. The RF Bluetooth/Wi-Fi signal received by the antenna is first amplified by the
LNA, and then mixed down with the quadrature output of the local oscillator (LO) to
baseband. The quadrature output of the mixer is then passed through a baseband channel
select LPF which has tunable bandwidth to suit both standards. The output of the LPF is
amplified by the variable gain amplifier (VGA). Since the maximum and minimum signal
levels are -4 and -80dBm (in Wi-Fi mode), RF front-end cannot handle the entire signal
range with only one LNA gain setting. Therefore, in addition to the VGA gain control,
103
the LNA has two gain settings, high gain and low gain modes. The LNA and VGA are
placed in an automatic gain control (AGC) loop to set the overall receiver gain depending
on the signal level at the receiver antenna. The signal level is measured in the digital
domain at the ADC output. Therefore, a digitally controlled VGA is desired.
reset
VGAwith offset cancellation
BT: 0/24dB, 1 stageWi-Fi: 0-62dB, 3 stage
Measuresignallevel
Gaincontrol
BT
Wi-Fi
BT
Wi-Fi
LPFBT: BW=600kHzWi-Fi: BW=6MHz ADC
BT: 11-bit 11MHzWi-Fi: 8-bit 44MHz
reset
Pipeline
Dig
ital M
UX
Pipeline
fref = 2MHz4.9GHz
2
2.45GHzI Q
LPF CP PFD
VCO
Phase SwitchingBlock
ProgrammableDivider
15/16 Prescaler
LNA
-15dB Attenuator
RF IN
Pipeline
Dig
ital M
UX
Pipeline
gain=15dB
Fig. 5.3 Dual-mode 802.11b/Bluetooth receiver
In Wi-Fi mode, the VGA gain has 2dB steps. Therefore, after AGC settling, the
signal level at the ADC input will be adjusted within a couple of decibels, thus relaxing
the ADC dynamic range requirement. There are two different preambles and headers
defined in Wi-Fi standard [3]: the mandatory supported long preamble and header, which
interoperates with the 1Mbit/s and 2Mbit/s DSSS specification (as described in IEEE Std
802.11, 1999 Edition), and an optional short preamble and header. The long and short
preamble durations are 144µs and 72µs, respectively. In both short and long preambles,
there is enough time to obtain fine gain adjustment in the VGA. On the other hand, in
104
Bluetooth mode, the preamble time is very short (4µs, which is equivalent to only 4 bits
long). Thus, settling time of the AGC loop has to be extremely small, not exceeding 4µs.
To alleviate this problem, only two VGA gain settings are used in Bluetooth mode. At
start up, the LNA and the VGA are set to the high and low gain modes, respectively.
According to the Bluetooth signal level, either the LNA is switched to the low gain mode,
the VGA is switched to the high gain mode, or the LNA and VGA gains are kept as they
are. Therefore, a maximum of only one gain adjustment step is made in Bluetooth mode.
As a result, the signal level at the ADC input can vary by about 30dB. This increases the
required ADC resolution by about 5 bits.
5.3. System Design Issues
In the following few subsections, system design issues in DCR architectures are
discussed [12]. The goal is to determine the optimal specifications for the system and for
each building block System and blocks specifications are derived in section 5.5. The
discussions are supported with simulation results that are performed using System View
and MATLAB. System View has a library for 802.11b standard that we used mainly to
examine the effect of the receiver non-idealities on the bit error rate (BER) performance
of the demodulator. Fig. 5.4 shows the basic simulation setup for the Wi-Fi simulations
using System View. On the other hand, MATLAB is used to assess the Bluetooth
demodulator performance in the presence of these non-idealities. A MATLAB code
(appendix B) was developed to simulate the effects of the receiver non-idealities on the
BER performance of the Bluetooth GFSK optimal non-coherent detector. Block diagram
of the Bluetooth GFSK demodulator is depicted in Fig. 5.5. The impulse response of the
105
Gaussian filter used in Bluetooth extends from –Tb to Tb (Tb is the bit duration) as shown
in Fig. 2.3. This means the intersymbol interference (ISI) occurs with the previous and
the following bits. To make the demodulator immune against the effect of ISI, eight
matched filters are used for all the combinations of the three bits (the current, previous,
and the following bits).
5.3.1. BER versus SNR
In order to evaluate the effect of each of the receiver non-idealities, we need to know
the minimum SNR to meet the required BER in the ideal case for each standard. In this
simulation, the signal is assumed to be perfectly synchronized with the demodulator
correlator. Since a digital demodulator is used for each mode, the correlator performance
depends on the sampling rate. Higher sampling rate means better demodulator
performance, but the simulation time is longer. To have a reasonable accuracy of the
simulation results compared to the ideal case, the sampling rates used are 20MSample/s
and 88MSample/s for Bluetooth and Wi-Fi modes, respectively. Figs. 5.6 and 5.7 show
the ideal case BER-SNR curves for Bluetooth and Wi-Fi modes where it is seen that the
required SNRmin is 12.25 and 11.4dB, respectively.
106
Fig. 5.4 Basic Wi-Fi simulation setup in System View
Select theMaximum
s*1(t)
r(t)
DetectedBit Stream
d
s*2(t)
s*8(t)
))((0∫T
dtRe
))((0∫T
dtRe
))((0∫T
dtRe
nT
Fig. 5.5 Bluetooth GFSK non-coherent demodulator used in MATLAB
107
10 10.5 11 11.5 12 12.5 13 13.5 1410
-4
10-3
10-2
BE
R
SNR
BER Performance of Optimal Non-coherent G FSK Demodulator
Fig. 5.6 BER versus SNR in Bluetooth mode in the ideal case (Sampling rate = 20MSample/s)
10 .4 10 .6 10 .8 11 11 .2 11 .4 11.6 11 .8 1210
-6
10-5
10-4
BE
R
SNR
Fig. 5.7 BER versus SNR performance of Wi-Fi 11Mbit/s CCK demodulator (Sampling rate = 88 MSample/s)
108
5.3.2. DC Offset Problem
Since in DCR architecture the downconverted band extends to zero frequency,
extraneous offset voltages can corrupt the signal, and more importantly, saturate the
following stages. To understand the origin and impact of offsets, consider the receiver
shown in Fig. 5.8, where the LPF is followed by an amplifier and an analog-to-digital
converter (ADC). Let us make two observations.
ADC
reflectionsfrom surrounding
objects
A B C X
LOleakage LO
VGA
LPF
(a)
ADCA B C X
Interfererleakage
LO
VGA
LPF
(b)
Fig. 5.8 Self-mixing of (a) LO (b) Interferers
109
First, the isolation between the LO port and the inputs of the mixer and the LNA is
not perfect. In other words, finite amount of feedthrough exists from the LO port to
points A and B (Fig. 5.8(a)). This effect is called “LO leakage” and it arises from the
capacitive and substrate coupling, and if the LO signal is provided externally, the leakage
comes from the bond wire coupling. The leakage signal appearing at the inputs of the
LNA and the mixer is now mixed with the LO signal, thus producing dc offset at point C.
This phenomenon is called “self mixing”. A similar effect occurs if a large interferer
leaks from the LNA or mixer input to the LO port and is multiplied by itself (Fig. 5.8(b)).
Second, the total gain from the antenna to point X is typically around 80-100dB so as
to amplify the microvolt input signal to a level that can be digitized by a low cost, low
power ADC. Of this total gain, typically 25-35dB is contributed by the LNA/mixer
combination. The offset at the mixer input due to LO coupling to the LNA input can be
expressed as:
Output offset ( ) LNA MixerLO power LO to LNA coupling G G= + + + (5.1)
With the above observations, we can obtain a rough estimate of the offset resulting
from self-mixing to appreciate the problem. Suppose in Fig. 5.8(a) the LO has a peak-to-
peak swing of 0.63V (≈ 0dBm for R = 50Ω)3 and experiences an attenuation of 60dB as it
couples to point A. If the gain of the LNA/ mixer is 30dB, then the offset produced at the
output of the mixer is -30dBm, which is equivalent to 10mV. We also note that the
desired signal level at this point can be as low as 30µVrms. Thus, if directly amplified by
VGA gain (could be up to 50-60dB), the offset voltage saturates the VGA stages.
3 To convert from volts to dBm, we use the formula: ( ) ( ) 30P indBm P in dBW= + , where
2
( ) 10log( )8
P PVP in dBWR
−=
110
The offset problem is exacerbated if self-mixing varies with time. This occurs when
the LO leaks to the antenna and is radiated and subsequently reflected from moving
objects (e.g. a car moving at high speed) back to the receiver.
DC offset can also be generated due to mismatches between circuit components.
Typical CMOS transistor VT mismatch can be in the order of few millivolts. Therefore,
Offsets generated by the mixer and filter can also saturate the VGA stages.
We infer from the above discussion that DCR architecture necessarily requires some
means of DC offset removal. Depending on the DC offset removal approach, residual dc
offset might appear at the ADC input. If not removed in the digital domain, this DC offset
may cause degradation in BER performance of the Bluetooth and Wi-Fi demodulator.
Figs. 5.9 and 5.10 show the demodulator degradation due to DC offsets in Bluetooth and
Wi-Fi modes, respectively. In order to have less than 0.3dB degradation in the required
SNR, the residual offset at the demodulator input should be less than 10% (relative to the
peak signal value) in Bluetooth and 5% in Wi-Fi.
111
9 9.5 10 10.5 11 11.5 12 12.5 13 13.5 1410
-4
10-3
10-2
BE
R
SNR
25%20%
10%5%
0%
15%
Fig. 5.9 BER performance of Bluetooth GFSK demodulator for different DC offsets (percentage relative to the peak signal value)
10.5 11 11.5 12 12.5 13 13.5 1410
-6
10-5
10-4
BE
R
SNR
15%
10%
5%
0%
Fig. 5.10 BER performance of Wi-Fi 11Mbit/s CCK demodulator for different DC offsets (percentage relative to the peak signal value)
112
It should be mentioned that no offset compensation scheme is implemented in the
demodulator for the simulations results shown in Figs. 5.9 and 5.10. Degradation due to
dc offset can be reduced by digitally compensating the residual offset. Two possible
approaches to eliminate the dc offset are discussed in the next two subsections.
5.3.2.1. AC coupling
A possible approach to remove the dc offset is to employ, i.e., passive RC high-pass
filtering, in the downconverted signal path [40]. However, the spectra of the Bluetooth
and Wi-Fi baseband signal exhibit a peak at dc as shown in Figs. 5.11 and 5.12. Such
signals are corrupted if filtered with a high corner frequency. Simulations using
MATLAB and System View were conducted to measure the BER performance
degradation due to AC coupling. As will be described in section 6.4, four first-order
HPFs are used in the VGA, one in front of each of the three VGA stages and one at the
output of the last stage. In Bluetooth mode, since only one VGA stage is used, the signal
is passed through the last two HPFs. On the other hand, since there are three VGA stages
used in Wi-Fi mode, the signal is passed through the four HPFs. The effects of these
HPFs were included in the simulations. Fig. 5.13 shows the Bluetooth GFSK
demodulator BER simulation using MATLAB with two cascaded HPF with the same
cutoff frequency. The degradation is less than 0.1dB if the cutoff frequency is less than
1kHz. In Wi-Fi CCK demodulator simulation using System View with four cascaded
HPFs. The cutoff frequency of the last two HPFs is set to 1kHz as required in the
Bluetooth mode. The cutoff frequency of the first two HPFs is swept in the simulation.
Fig. 5.14 shows that the degradation in CCK demodulator performance is less than 0.2dB
if the cutoff frequency of the first two HPFs is less than 5kHz.
113
Fig. 5.11 Power spectral density of Bluetooth GFSK signal
0 5 10 15 20 25 30 35 40 45-120
-100
-80
-60
-40
-20
0
Frequency (MHz)
CCKpowerspectrumdensity
Fig. 5.12 Power spectral density of Wi-Fi CCK signal
114
10 10.5 11 11.5 12 12.5 13 13.5 1410
-4
10-3
10-2
BE
R
SNR
3kHz
2kHz
1kHz
noHPF
cutoff ofthe twoHPFs
Fig. 5.13 Bluetooth GFSK demodulator performance with two HPFs
10.5 11 11.5 12 12.510
-6
10-5
10-4
BE
R
SNR
15kHz
10kHz
5kHz
noHPF
cutoff of 1st
two HPFs
cutoff of lasttwo HPFs is
1kHz
Fig. 5.14 Wi-Fi 11Mb/s CCK demodulator performance with four HPFs
115
Another issue concerning using passive RC HPF to remove the offset is the thermal
noise of the resistor R. The integrated thermal noise at the output of the HPF is 2kT/C
(the factor 2 is due to differential operation), where k is Boltzmann constant and T is the
absolute temperature. If the received signal is at -85dBm (12.6µVrms), and assuming the
total gain of the LNA, mixer, and LPF is 37dB, the differential signal level at the VGA
input will be -85 + 37 = -48dBm (0.89mVrms). If the first AC coupling stage is placed at
the VGA input, the two differential capacitors must at least 1pF each so that 2kT/C
remains 20dB below the signal level. AC coupling capacitors at later VGA stages can be
smaller because the signal level is larger. The low frequency part of the kT/C noise
spectrum of the first HPF stage is partially rejected by the later HPF stages.
5.3.2.2. Offset cancellation loop
By sensing the offset at the ADC input, integrating it, and then injecting the
correction signal at the input of the VGA (a point where the offset is not too high to affect
the circuit operation) as shown in Fig. 5.15 [41, 42]. If the gain of the VGA is K, the
transfer function of the closed loop system shown in Fig. 5.15 is expressed as:
Ks
sKK
s
K
oo ωω +=
+=
1response loop Closed (5.2)
Therefore, the overall response of the closed loop VGA with the offset cancellation
feedback loop is a HPF response and, like AC coupling, it still corrupts the signal
spectrum at low frequencies. Note that in equation (5.2), the HPF cut-off frequency
depends on the integrator time constant and the loop gain (including the VGA gain).
Therefore, the cutoff frequency of the HPF response will vary with the VGA gain, unless
this gain variation is compensated in the feedback gain such that Koω is kept constant.
116
Thus, this approach is more complex than AC coupling. Alternatively, as depicted in Fig.
5.16, the DC offset can be sensed in the digital domain at the output of the ADC and,
therefore, requiring a digital-to-analog converter (DAC) to inject the correction signal at
the VGA input [43]. Unlike pure analog approaches, the injected correction signal is
available digitally at the DAC input and it can be stored forever. Therefore, once the
proper correction digital word at the DAC input is found, the loop can be opened and the
correction digital word is stored and applied to the DAC input. The main advantage of
this approach is that the signal is dc coupled through the VGA and the low frequency
portion of the signal spectrum is not corrupted. However, this approach is rather more
complex than AC coupling or analog offset cancellation loop. It requires high DAC
resolution (resolution increase with VGA maximum gain). In addition, the DAC output
swing has to be as low as the dc offset range at the injection point which is usually which
is usually in the order of tens of millivolts.
Offset cancellation scheme using ac coupling HPFs has been used in the dual mode
receiver. The scheme is described in detail in chapter VI.
ADC
soω
offset cancellationloop using integrator
K(s)
Fig. 5.15 Analog offset cancellation loop using feedback integrator [41, 42]
117
ADC
offset cancellationloop using DAC
DAC
offset correctiondigital word
Fig. 5.16 Offset cancellation loop using DAC [43]
5.3.3. Filter Frequency Response
Several simulations have been done to assess the effect of filter approximation and
bandwidth on the BER performance demodulator in both modes. Fig. 5.17 shows the
Bluetooth demodulator performance using 5th order Butterworth LPF and 4th Chebychev
LPF with 0.5dB pass-band ripples. Both filters have 600 kHz bandwidth. There is more
than 0.5dB degradation when using Chebychev approximation compared to Butterworth
approximation. This can be attributed to the fact that Chebychev filter has more group
delay variation in the pass-band than in Butterworth. Therefore, a Butterworth
approximation is chosen for this receiver. The next step is to determine the optimum filter
bandwidth. The higher the filter bandwidth is, the less attenuation for adjacent channel
interferer and the more noise passed to the demodulator. The lower the bandwidth is, the
more significant portion of the signal spectrum is rejected. Fig. 5.18 shows the Bluetooth
BER performance using Butterworth filter with different bandwidths. The optimum
Butterworth filter bandwidth is 600kHz in Bluetooth mode.
118
In Wi-Fi mode, on the other hand, 5th order Butterworth filter with a bandwidth of
6MHz has superior BER performance over a 4th order Chebychev with the same
bandwidth as depicted in Fig. 5.19. Fortunately, 5th order Butterworth is the optimum
filter approximation for both standards. Therefore, there is no need to change the relative
poles locations when switching between the standards and only frequency tuning is
required. The optimum Butterworth filter bandwidth is obtained from the simulation
plotted in Fig. 5.20. 6MHz appears to be the bandwidth corresponding to best BER
performance. However, simulations showed that the BER performance does not degrade
significantly when the cutoff frequency is changed by ±500 kHz; relaxing the filter
frequency tuning accuracy.
BER Perform ance of Optimal N on-coherent GFSK Demodulatorwith Butterworth and C hebychev LPF
10 10.5 11 11.5 12 12.5 13 13.5 1410
-4
10-3
10-2
BE
R
SNR
5th order Butterworth600kHz
4th order Chebychev600kHz, 0.5dB ripples
Fig. 5.17 BER performance using different LPF approximations in Bluetooth mode
119
10 10.5 11 11.5 12 12.5 13 13.5 1410
-4
10-3
10-2
BE
R
SNR
400kHz
500kHz
600kHz
700kHz
no LPF
BER Performance o f Optimal Non-coherent GFSK Demodulatorfor d ifferent LPF bandw idths
Fig. 5.18 Bluetooth BER performance using Butterworth LPF with different bandwidths
11.5 12 12.5 13 13.510
-6
10-5
10-4
BE
R
SNR
4th orderChebychev
5th orderButterworth
Fig. 5.19 Wi-Fi 11Mbit/s CCK BER using different LPF approximations
120
10 10.5 11 11.5 12 12.5 13 13.5 1410
-4
10-3
10-2
BE
R
SNR
400kHz
500kHz
600kHz
700kHz
no LPF
BER Perform ance of Optim al Non-coherent GFSK Demodulatorfor different LPF bandw idths
Fig. 5.20 Wi-Fi BER performance using Butterworth LPF with different bandwidths
5.3.4. ADC Resolution and Sampling Rate
To avoid degrading the SNR, the ADC quantization noise should be well below the
noise floor at the VGA output. In other words, the ADC dynamic range should be higher
than the required SNRmin which is about the same in both modes. Fig. 5.21 shows the
degradation due to ADC quantization in Bluetooth mode. The degradation is about 1dB
for 3bit ADC, which is expected since it has only 18dB of dynamic range, not sufficiently
higher than SNRmin. With 4bit quantization, degradation is only 0.2dB degradation
relative to the ideal case with no quantization. But the signal level at the ADC input can
vary by about 26dB. To cover this variation, the ADC resolution should be increased by
about 5bits. So the minimum required ADC resolution is 9bits in Bluetooth mode. To
allow for some margin for implementation loss and signal dynamic range, 11bits ADC is
121
designed. Fig. 5.22 shows the effect of the sampling rate on the BER performance of the
Bluetooth demodulator. A lower sampling rate leads to higher timing errors between the
incoming signal and the demodulator matched filter and, therefore, greater loss in
SNRmin. Fig. 5.22 shows that the degradation is less than 0.1dB for sampling rates of
8MHz or higher. A sampling rate of 11MHz is used in the Bluetooth mode. This
frequency is chosen because it can be derived easily from the 88MHz crystal oscillator.
In Wi-Fi mode, the degradation due to ADC quantization is depicted in Fig. 5.23.
The degradation is less than 0.1dB for 5 bits of resolution. However, 8bits are
implemented in Wi-Fi mode to account for AGC resolution, possible signal level
variations, error in signal level measurement. Fig. 5.24 shows the effect of ADC
sampling speed on the Wi-Fi BER performance. 44MHz appears to be a good
compromise between ADC power consumption and BER degradation.
10.5 11 11.5 12 12.5 13 13.5 14 14.510
-4
10-3
10-2
BE
R
SNR
3bit
4bit
5bit
6bit
noquant
Fig. 5.21 Bluetooth GFSK BER performance for different ADC quantization bits
122
10 10.5 11 11.5 12 12.5 13 13.5 1410
-4
10-3
10-2
SNR
BE
R
4MHz
6MHz
8MHz
10MHz
40MHz LPF BW = 600kHz
Fig. 5.22 Bluetooth GFSK BER performance for different ADC sampling rates
10.5 11 11.5 12 12.510
-6
10-5
10-4
BE
R
SNR
4bits
5bits
6bits
noquantization
Fig. 5.23 ADC quantization effect on BER performance of Wi-Fi 11Mb/s CCK
123
9.5 10 10.5 11 11.5 12 12.5 13 13.510
-6
10-5
10-4
SNR
BE
R
LPF BW = 600kHz
22MHz
44MHz
88MHz
Fig. 5.24 Wi-Fi 11Mb/s CCK BER performance for different ADC sampling rates
5.3.5. VGA Dynamic Range
The dynamic range of the VGA is determined by the signal power level range at the
receiver antenna. In Wi-Fi, the minimum (sensitivity) and maximum signal levels at the
receiver antenna are specified to be -80dBm to -4dBm for the 2Mbit/s data rate. To cover
this 76dB variation in the signal level, the overall gain of the receiver must be adjusted
such that, for all signal levels, no block is saturated and the signal level at the ADC input
is within the allowable range to demodulate the signal. To avoid saturating the RF front-
end at high signal levels, the LNA is bypassed and an attenuator is used. The
LNA/attenuator stage provides 30dB gain step. Therefore, reducing the signal level
dynamic range to 76-30 = 46dB at the blocks following the LNA. In Wi-Fi, signal level
at the input of the ADC has to be well defined to make the most out of the 8-bit ADC
resolution. A VGA with 2dB gain step is used to reduce the signal dynamic range at the
LPF BW = 6MHz
124
ADC input to 2dB. To have some margin in sensitivity and maximum signal levels
specifications, a 0-62dB VGA is designed with a total of 16dB margin at the high and
low ends. This large margin is also used to compensate for possible gain errors in the
receiver blocks due to process variations and non-ideal components, especially in the RF
front end, and to achieve better sensitivity and maximum signal levels.
In Bluetooth mode, due to the very short time allowed for the preamble, the settling
time of the AGC loop has to be extremely short. Therefore, only two VGA gain settings
are used (0/24dB using the 3rd VGA stage only). Upon reception of the Bluetooth
preamble, the LNA (VGA) is set to maximum (minimum) gain; then, a maximum of one
gain control step is needed to adjust the receiver gain, minimizing the overall receiver
settling time. Due to the large gain step employed in Bluetooth mode, the signal level at
the output of the VGA may vary by about 30dB, and the ADC must be designed to
accommodate this dynamic range for the Bluetooth signal.
5.3.6. I/Q Mismatches
In low-IF receivers [15, 44], gain and phase mismatches between I and Q channels
affect the image rejection ratio of the receiver. Although the image problem is eliminated
in DCR architecture, the gain and phase mismatches still affect the BER performance of
the demodulator. However, matching requirements in DCR is not as tough as in low-IF
architecture. Effects of gain and phase mismatches on the Bluetooth BER performance
are depicted in Figs. 5.25 and 5.26, respectively. For less than 0.2dB loss in SNRmin, gain
and phase mismatches should be less than 10% and 5°, respectively. The Wi-Fi BER
degradations due to gain and phase mismatches are plotted in Figs. 5.27 and 5.28. SNRmin
degradation is below 0.1dB for gain and phase mismatches 5% and 5°, respectively.
125
9 9.5 10 10.5 11 11.5 12 12.5 13 13.5 1410
-4
10-3
10-2
BE
R
SNR
BER performance of optimal non-coherent GFSK demodulatorfor different gain mismatches
50%
20%
10%
5%
0%
Fig. 5.25 BER performance of Bluetooth demodulator for different gain mismatches
10.5 11 11.5 12 12.5 13 13.5 1410
-4
10-3
10-2
BE
R
SNR
0o
5o
10o
15o
20o
BER performance of optimal non-coherent GFSK demodulatorfor different phase mismatches
Fig. 5.26 BER performance of Bluetooth demodulator for different phase mismatches
126
10.5 11 11.5 12 12.510
-6
10-5
10-4
BE
R
SNR
30%
20%
10%
0%
Fig. 5.27 Wi-Fi 11Mb/s CCK BER performance for different I/Q gain mismatches
10.5 11 11.5 12 12.510
-6
10-5
10-4
BE
R
SNR
15o
10o
5o
0o
Fig. 5.28 Wi-Fi 11Mb/s CCK BER performance for different I/Q phase mismatches
127
5.3.7. Even-order Distortion
Typical RF receivers, with non-zero IF frequency, are susceptible to only odd-order
intermodulation effects because even order distortion results in harmonics outside the
band of interest. In direct conversion, on the other hand, even-order distortion becomes
problematic. Suppose there are two strong interferers )cos( 11 tA ω and )cos( 22 tA ω , close
to the channel of interest placed at LOω , experience a non-linearity such as
)()()( 221 txtxty αα += in the LNA. Then )(ty contains a 2nd order intermodulation term
(IM2) term: tAA )cos( 21212 ωωα − indicating that two high-frequency interferers generate
a low-frequency beat in the presence of even-order distortion. Upon multiplication of
tLOωcos in an ideal mixer, such a term is translated to high frequencies and hence
becomes unimportant. In reality, however, mixers exhibit a finite direct feedthrough from
the RF to the IF output. This feedthrough results from asymmetry in the circuit which is
caused by mismatches between transistors or the deviation of the LO duty cycle from
50%. Thereby, producing an output signal such as )cos)(( tAatv LORF ω+ . Thus a fraction
of )(tvRF appears at the output with no frequency translation. If )( 21 ωω − is small
enough, the IM2 term appears inside the signal band and cannot be rejected by the filter.
5.3.8. VCO Frequency Pulling
An interesting phenomenon may happen when a large signal is injected into the LO.
If the injected signal is close to the VCO frequency, the VCO may shift towards the
injected signal frequency and eventually “lock” to that frequency. Called “injection
locking,” this effect is described in [45, 46]. In transceiver environment, various sources
can introduce oscillator pulling. For example, the power amplifier (PA) output may
128
couple to the local oscillator. Another example of injection pulling arises in the receiver
path when the desired signal is accompanied by a large interferer as illustrated in Fig.
5.29. If the interferer frequency is close to the LO frequency, coupling through the mixer
may pull LOω toward intω . This problem is rather serious in Bluetooth since as the
interferer can be as large as -27dBm, and it will be amplified to -12dBm after the LNA.
To avoid LO pulling, the VCO is run at double the RF frequency, then a divide-by-
two is used to obtain the desired LO frequency. In this case, the VCO frequency is far
apart from the PA and the interferer frequencies.
LNA
ωο ωωint
InterfererDesiredsignal
Fig. 5.29 Injection pulling due to large interferer
5.3.9. Combined Effects
BER performances of the Bluetooth and Wi-Fi demodulators have been simulated
with all parameters and non-idealities mentioned in the previous subsections. The values
of the parameters and non-idealities used in the simulations are listed in table 5.1 for both
standards. Fig. 5.30 shows the BER performance of the Bluetooth demodulator in the
ideal case and when all effects of the values listed in table 5.1 are included. Note that the
required SNRmin becomes 13.6dB. Degradation due to the combined effects of the non-
idealities is about 1.4dB. Degradation due the combined non-idealities effects in Wi-Fi
mode is depicted in Fig. 5.31. The required SNRmin is 13.2dB, about 1.75dB above the
129
ideal case. These SNRmin values will be taken as reference in the determination of the
required dual mode receiver parameters in the next section, and therefore, all non-
idealities are taken into account in the estimation of the receiver performance.
Table 5.1 Receiver parameters and non-idealities used in the BER simulations
Bluetooth Wi-Fi
DC offset 10% 5%
HPF cutoff 1kHz 5kHz
5th order Butterworth LPF cutoff 600kHz 6MHz
ADC quantization 5bits 6bits
ADC sampling rate 10MHz 44MHz
I/Q gain mismatch 10% 10%
I/Q phase mismatch 5° 5°
10 11 12 13 14 15 1610
-4
10-3
10-2
BE
R
SNR
allnonidealities
ideal
Fig. 5.30 Effects of combined receiver non-idealities on Bluetooth BER performance
130
10.5 11 11.5 12 12.5 13 13.5 1410
-6
10-5
10-4
BE
R
SNR
allnonidealities
ideal
Fig. 5.31 Effects of combined receiver non-idealities on Wi-Fi BER performance
5.4. From Standard to Block Specifications
At this point, we have enough information about the two standards to start translating
standard specifications into receiver system specifications, and then to individual
building blocks specifications. Fig. 5.32 shows the flowchart of the entire receiver system
design process.
131
NF IIP3
IIP2
System levelspecifications
BER SystemLevel Simulations Adjacent
ChannelTest
Standard Specifications
Building blocksspecifications
NO
YES
Circuit level design
layout
NO
Meet systemspecs?
Meet specsand powerbudget?
post layoutsimulations
Meet specs?
ICfabrication
YES
NO
ICMeasurements
Meet specs?
YES
NO
YES
Massproduction
Fig. 5.32 Flowchart of the system design process
5.4.1. From Standard to Receiver Specifications
5.4.1.1. Standard receiver tests
In wireless receivers’ standards, a set of receiver tests are specified to verify if the
receiver complies with the standard. Among these tests, the most relevant are: sensitivity
test, adjacent channel test, and intermodulation test.
132
Sensitivity test: Receiver sensitivity is defined as the minimum input signal power
that is required to achieve a certain specified at the receiver output BER. In the sensitivity
test, only the desired signal is applied to the receiver input at the required sensitivity
level. The signal BER measured at the receiver output must be better than the specified
BER. This test is used to derive the receiver required NF.
Adjacent channel test: In this test, the desired signal is applied to the receiver at a
level, usually few decibels higher than the specified sensitivity level. In addition to the
desired signal, another signal is applied at the adjacent channel frequency, at a power
usually higher than the desired signal. Under these conditions, the BER measured at the
receiver output must be better than the BER specified by the standard. This test is used to
derive the receiver IIP2 and IIP3.
Intermodulation (IM) test: In this test, two high-power single (or modulated) tones
are applied to the receiver input in presence of the desired signal. The BER measured at
the receiver output must be better than the specified BER. Usually, 3rd order IM test
(IM3) is specified. In this case, the two tones are applied at frequencies fff ∆+= 01 and
fff ∆+= 202 , where 0f is the frequency of the desired channel and f∆ is the two tones
frequency separation. This test is usually used to determine the required receiver IIP3.
Maximum signal test: This test determines the maximum signal that can be applied
to the receiver while still meeting the BER requirement. The difference between the
maximum signal level and sensitivity level sets the receiver dynamic range. To have a
well-defined signal level at the demodulator at the end of the receiver chain, the receiver
gain has to be changed according to the signal level. Thus, sensitivity and maximum
signal levels determine the maximum and minimum gain of the receiver.
133
5.4.1.2. Obtaining receiver key specifications
The key specifications of a direct-conversion receiver are the NF, IIP3, IIP2. These
specifications are derived from the standard tests described above.
5.4.1.2.1. Noise figure (NF)
The noise floor of the receiver is expressed as:
NFBWNFBW ++−=++= )10log(174)10log(log(KT)10floorNoise (5.3)
where K is Boltzmann constant, T is the absolute temperature (290°K is usually
used), NF is the receiver noise figure, and BW is the channel selection filter bandwidth.
To achieve the specified BER, certain SNRmin must be achieved. The SNR-BER curve is
obtained from baseband system level simulations as described in section 5.4.1. The
receiver sensitivity can be expressed in terms of SNRmin and noise floor as follows:
marginfloor noiseySensitivit min ++= SNR (5.4)
Several dBs of margin are taken to account for possible degradation due to process
variations, mismatches, or other effects that are unaccounted for in the above equation.
Therefore, the required receiver NF can be expressed as:
margin))log(10174(ySensitivit min −+−−−= BWSNRNF (5.5)
5.4.1.2.2. Third order intercept point (IIP3)
The IIP3 is derived from the two tone test. We can roughly state that the IM3 product
must be less than the desired signal by the required SNRmin. In other words, we are
assuming that the effect of the IM3 product is similar to the noise effect. Although this is
a rough approximation, it gives a good idea about the required system IIP3 without the
134
need to run the actual two tone simulation. Such a simulation can take very long time
because of the two higher frequency tones used.
The input referred IM3 product can be written as:
minint3 323 SNR-PIIPPP sigIIM <−= (5.6)
Where intP and sigP are the interferer and signal levels when the IM test is conducted.
Therefore, the minimum receiver IIP3 is given by:
margin3213 minint ++−= )SNRPP(IIP sig (5.7)
Again, several dBs of margin are added to account for implementation loss.
5.4.1.2.3. Second order intercept point (IIP2)
Similar to the IIP3 test, this test is carried out using two nearby interferers at 1f and
fff ∆+= 12 . Due to the even-order distortion of the receiver, a low-frequency beat at
fff ∆=− 12 appears at the receiver output. Such a beat signal has to be less than the
desired signal by the SNRmin value, therefore:
minint2 22 SNR-PIIPPP sigIIM <−= (5.8)
Thus, the minimum receiver IIP2 is given by:
margin 22 minint ++−= SNRPPIIP sig (5.9)
5.4.1.3. Bluetooth receiver specifications
NF: The required Bluetooth receiver sensitivity is -70dBm. Baseband MATLAB
simulations showed that, to achieve 10-3 BER, an SNRmin of 13.6dB is required. For
600kHz channel selection filter bandwidth, and using 6dB of margin, the required system
NF is calculated by substituting these values in equation (5.5):
135
dBNF 6.266))10600log(10174(6.1370- 3 =−×+−−−= (5.10)
This is a very relaxed number for the system NF. Many implementations [44, 47-49]
have achieved sensitivities better than -80dBm. Since the required Wi-Fi NF is much
tougher, it is reasonable to shoot for about -84dBm sensitivity for Bluetooth. This target
sensitivity is 14dB above the specified -70dBm. Therefore, the corresponding receiver
NF is 14dB less than the value obtained in equation (5.10). This gives a target NF of
12.6dB.
IIP3: In the intermodulation test, the receiver has to meet the 10-3 BER when the
following signals are applied:
• The wanted signal at frequency of with a power level -64dBm.
• A static sine wave signal at 1f with a power level of –39dBm.
• A Bluetooth modulated signal at 2f with a power level of -39dBm.
Such that 212 fffo −= ( 1f is closer to the signal than 2f ) and MHznff 112 ×=− ,
where n can be 3, 4, or 5. The system must fulfill one of these alternatives.
Substituting these values in equation (5.7), using 6dB margin:
dBm)(IIP 7.1366.13)64()39(3213 −=++−−−= (5.11)
IIP2: There is no specified IIP2 test in the Bluetooth standard. However, the IIP2
can be derived from the adjacent channel test or the intermodulation test, whichever
requires higher IIP2. The largest specified interferer is -27dBm Bluetooth modulated
signal at offset frequency ≥ 3MHz, where the signal is at dBm67− . Since the interferer is
frequency modulated (no amplitude modulation), 2nd order distortion in the receiver will
result in dc offset. Unfortunately, AC coupling does not reject this dc offset because the
136
LNA input is grounded during the dc offset estimation (to avoid corrupting dc
information by the incoming signals). However, this offset can be rejected during the
synch word before receiving the actual data. Therefore, the dc offset does not have to be
much smaller than. This offset has to be lower than the signal by SNRmin = 13.6dB. This
can be viewed as if we have two very close interferers, with power dBm30327 −=−−
each. Therefore from equation (5.9):
dBmIIP 6.2666.13 )67()30(22 =++−−−= (5.12)
5.4.1.4.Wi-Fi receiver specifications
NF: The sensitivity level at 11Mbit/s data rate is -76dBm with a frame error rate
(FER) of 0.08. The frame length is 1024 octets (or equivalently 8×1024 = 8092bits). So
this FER corresponds to a BER of 0.08/8092 ≈ 10-5. Based on System View Baseband
simulations, the SNRmin to achieve 10-5 BER is 13.2dB. If the channel selection filter
bandwidth is 6MHz, and using 6dB of margin, the required system NF from equation
(5.5) is:
dBNF 116))106log(10174(2.1376- 6 =−×+−−−= (5.13)
In this design, our target sensitivity is -80dBm, 4dB above the Wi-Fi sensitivity
specification. Therefore, the required NF is 11-4 = 7dB.
IIP3: There is no standalone IIP3 test in Wi-Fi standard. However, the required IIP3
can be estimated from the adjacent channel test. In this test, an 11Mbit/s Wi-Fi signal is
applied at 25MHz away from the desired signal. The desired and interference signal
levels are -70dBm and -35dBm, respectively. Since the interferer is a wideband signal,
direct application of equation (5.7) is not accurate. A MATLAB program is developed
(appendix C) to compare the intermodulation products of the CCK signal at 25MHz away
137
from the signal with the intermodulation product of two tones at 16.7 and 33.4MHz at the
same power level. If the 3rd harmonic product is passed through a 5th order 6MHz filter,
the CCK IM3 output will be 37dB less than the two tones IM3 product. Therefore, in
order to calculate the required system IIP3, equation (5.6) will be modified as follows:
SNR-PIIPP sig 37323 int <−− (5.14)
Then the minimum IIP3 in Wi-Fi mode is:
dBm
SNRPPIIP sig
4.236)2.137037)35(3(21
margin) 373(213 minint
−=+++−−×=
++−−= (5.15)
It is worth mentioning that the IM3 product of the CCK signal is obtained by
convolving the magnitude of the CCK spectrum with itself three times. This implies that
the obtained IM3 product is actually a worst-case number. In other words, the required
IIP3 calculated in (5.15) is rather a conservative value.
IIP2: Again, there is no IIP2 test in Wi-Fi standard. Therefore, we will follow the
same approach used in IIP3 calculations to calculate the IIP2 using the adjacent channel
test. A MATLAB code (appendix C) is developed to compare the IM2 product of the
CCK signal with the IM2 in the standard IIP2 test using two tones at 24 and 26MHz. The
CCK IM2 product at the output of a 6MHz 5th order Butterworth filter is 20dB less than
the two tone IM2 product. Therefore, equation (5.8) can be modified as follows:
SNR-PIIPP sig 2022 int <−− (5.16)
Then the minimum required IIP2 is estimated as:
dBmSNRPPIIP sig 8.0613.2(-70)-20-2(-35)margin 2022 minint −=++=++−−= (5.17)
A summary of required receiver specifications for both modes is listed in table 5.2.
138
Table 5.2 Summary of receiver specifications
Wi-Fi BluetoothNF (dB) 7 12.6
IIP3 (dBm) -23.4 -14.4IIP2 (dBm) -0.8 26.6
SpecificationsValue
5.4.2. Determination of Receiver Building Blocks Specifications
From the receiver system specifications, the specifications of the individual building
blocks can be obtained. Conversion from system to blocks specifications is rather an
iterative process. First we start with a reasonable distribution of specifications based on
previous experience and/or some intuition. Then the overall system specifications are
calculated to see if the system specifications are met. The distribution of blocks
specifications is subject to the restriction of sharing the same RF front-end (LNA, mixer,
and PLL) from both standards to save silicon area. In other words, the specifications of
these blocks will be set by the toughest requirement of both standards. Although a set of
key receiver specifications (NF, IIP3, and IIP2) were obtained in the previous section, we
still need to look at the standard tests, from time to time, to make sure that the receiver is
operating properly under all conditions. A MATLAB code (appendix D) is developed to
calculate the system parameters from the blocks specifications. This program is used to
check if the systems specifications are met for a set of blocks specifications.
5.4.2.1. Gain distribution
The first step to determine the specifications of the receiver blocks is to determine
the gain distribution. In order to relax the NF of the baseband blocks (filter, VGA, and
ADC), RF front-end gain should be maximized. The maximum RF front gain is
determined by the adjacent channel test. In Bluetooth mode, the largest adjacent channel
139
interferer is -27dBm. If the maximum signal at the output of the mixer is 4dBm (0.5Vp),
the total RF front-end (LNA+mixer) gain is 31dB. In the Wi-Fi case, the adjacent channel
power level is -35dBm, which is lower than Bluetooth interferer, and therefore, will not
saturate the mixer output. Since the interferer level is largest at the filter input, filter
linearity is very critical. An odd order filter is preferred because of the availability of a
passive pole that can be placed at the filter input to attenuate the interferer, and therefore,
improve the out-of-band linearity of the filter. A 5th order Butterworth filter is
implemented. To further relax the NF of the filter and the VGA, a 6dB gain is
implemented at the filter’s first biquad (after the passive pole). The maximum signal
swing at the ADC input is 2Vp-p. To allow for some margin, the total receiver gain is
adjusted such that the signal level at the ADC input is 1Vp-p (4dBm). To achieve Wi-Fi
sensitivity of -85dBm, the overall maximum receiver gain is 89dB. Therefore, the
maximum VGA gain is dB5263189 =−− . In Wi-Fi mode, the maximum signal is -
4dBm for the 2MHz data rate. This level is too high for the RF front-end and will saturate
the LNA. For such high signal levels, LNA is bypassed and the signal is passed through
an attenuator (low gain mode). To avoid saturating the filter and VGA, the attenuator
gain is chosen such that the signal level at the output of the filter is below 4dBm. If the
RF front-end low gain is 1dB (30dB drop from the high gain mode), the signal level at
the output of the filter is 3dBm, below the maximum level. To get a 4dBm signal level at
the ADC input, the minimum VGA gain is 1dB. According to typical values, the LNA
and mixer gains are chosen to be 15 and 16dB, respectively. Therefore, to have a 30dB
gain drop from high gain to low gain, the attenuator provides 15dB of attenuation. Table
5.3 shows the gain distribution of the receiver blocks. To cover for all possible errors in
140
the gains of the blocks, especially the RF blocks, the VGA gain is designed for a gain
range of 0 to 62dB.
In Bluetooth mode, due to the short preamble time, the overall receiver gain has to be
adjusted using the minimum number of steps. Changing the front-end gain between high
and low gain modes is unavoidable due to the large maximum signal level in Bluetooth (-
20dBm). If we shoot for sensitivity and maximum signal levels of -90dBm and -10dBm,
signal dynamic range at the receiver input is 80dB. Using 30dB gain step in the front-end,
the signal level at the VGA input can vary by about 50dB. If the VGA gain is fixed in
Bluetooth mode to minimize receiver gain settling time, signal dynamic range at the ADC
input will be 50dB, requiring more than a 13bits ADC to cover such dynamic range.
Therefore, only one large gain step is used in the VGA to reduce the signal dynamic
range at the ADC input. At maximum signal input (-4dBm in Wi-Fi mode), the LNA is
bypassed by the attenuator and the VGA gain is set to 0dB. At the sensitivity level, -
90dBm, the front-end is set to the high gain mode and the VGA gain is set at 24dB at
which the signal level at the ADC input is -29dBm. Thus, in Bluetooth mode, the signal
dynamic range at the ADC input is 80-30-24=26dB, requiring about 10bits of ADC
resolution which is a reasonable value for the Bluetooth data rate.
Table 5.3 Receiver gain distribution
LNA/attenuator Mixer Filter VGA (Wi-Fi/BT)
Maximum gain (dB) 15 16 6 62/24
Minimum gain (dB) -15 16 6 0
141
5.4.2.2. NF distribution
Starting with the gain distribution in table 5.3, we can distribute the NF among the
receiver block such that the required overall system NF is met. The overall system NF is
then calculated using the Friis equation [50] for cascaded stages:
21
22
21
22
21
22
1
21
111)1(1−
−++
−+
−+−+=
m
mtot AAA
NFAA
NFA
NFNFNFL
L (5.18)
Where NFi and Ai are the NF and gain of the ith stage, where the LNA is the first
stage. The Friis equation indicates that the noise contributed by each stage decreases as
the gain of the preceding stages increases, implying that the first few stages in a cascaded
are the most critical. In the case of our particular receiver, since the RF front-end gain is
limited by the adjacent channel test, NFs of LNA, mixer, and filter dominate the overall
system NF. Table 5.4 lists the NF of the receiver building blocks. Percentage
contributions to the overall system NF in Bluetooth and Wi-Fi modes are shown in the
pie diagrams in Fig. 5.33 and 5.34, respectively.
Table 5.4 NF, IIP3, and IIP2 distribution in Bluetooth/Wi-Fi mode
LNA Mixer Filter VGA ADC System
NF (dB) 3/3 20/15 36/32 30/30 57/65 9.49/6.46
IIP3 (dBm) -8 5 23 10 10 -13.5
IIP2 (dBm) 11 48 64 31 41 26.7
142
LNA
Mixer
Filter
VGAADC
12.6%
39.7%
40.1%
5%2.5%
Fig. 5.33 Noise contributions from the different receiver blocks in Bluetooth mode
LNA
Mixer
Filter
VGAADC
19.4%
32.1%
41.7%
6.6%
Fig. 5.34 Noise contributions from the different receiver blocks in Wi-Fi mode
5.4.2.3. IIP3 distribution
Since the interferer is filtered out by the LPF, the IIP3 of the receiver will be
dominated by the LNA, mixer, and LPF. The IIP3 of the RF front end will be the same
for both Bluetooth and Wi-Fi. Furthermore, it will be easier to design the filter if it has
143
the same IIP3 in both modes. Therefore, the IIP3 specification will be determined by the
tougher of the two standards which is -14.4dBm in Bluetooth mode. The IIP3 is
distributed among the LNA, mixer, and LPF as shown in table 5.4. Contribution of the
three blocks to the system IIP3 is depicted in Fig. 5.35. The overall system IIP3 is
calculated from the IIP3 of the individual blocks using the following equation:
23
21
22
21
23
22
21
23
21
23
23 311
11
mtot IIP
m
IIPIIPIIPIIP AAAA
AAA
AA
AA−++++=
LL (5.19)
Where iIIPA 3 and
iiA is the input referred IIP3 (in volts) and the linear voltage gain of
the ith block.
LNA
Mixer
Filter
28%
44%
28%
Fig. 5.35 IIP3 contributions from receiver blocks in both modes
5.4.2.4. IIP2 distribution
Similar to the IIP3, the system IIP2 is typically dominated by the mixer and the filter.
Ideally, all low-frequency components created by the even-order distortion on the LNA
144
gets upconverted by the mixer, and consequently filtered out by the LPF. However, If a
single-ended mixer is used, RF-IF isolation is poor and the even-order harmonic at the
output of the LNA propagate through the mixer, corrupting the downconverted signal
spectrum. Therefore, the IIP2 of the LNA contributes to the system IIP2 if a single-ended
mixer is used. For differential mixers, RF-IF isolation is better (typically around 30dB)
and it depends on matching between transistors. Differential double balanced mixer is
used in this design receiver to have good RF-IF and LO-RF isolation. LO-RF isolation is
important to alleviate the self-mixing problem discussed in section 5.4.2. The IIP2 is
distributed as shown in table 5.4. Contributions of the receiver blocks to the system IIP2
are depicted in Fig. 5.36. The following equation is used to calculate the system IIP2
based on the IIP2 of the individual blocks.
mtot IIP
m
IIPIIPIIPIIP AAAA
AAA
AA
AA 2
121
2
21
2
1
22 321
11 −++++=L
L (5.20)
Mixer
Filter
LNA+VGA+ADC
48% 48%
4%
Fig. 5.36 IIP2 contributions from receiver blocks in both modes
145
5.4.2.5. LO phase noise
Ideally, the output of the LO is a single tone signal at the frequency of the desired
signal (in DCR architecture). However, due to thermal and 1/f noise of the
transistors/resistors used in the PLL, amplitude and phase noise components appear at the
output of the LO. Amplitude noise is not critical because of the hard switching of the
mixer transistors. Phase noise, on the other hand, is very critical especially in the
presence of strong out-of-band interferers. This is due to the phase noise of the LO
modulating the carrier of the strong signal. The carrier is spread in frequency by the
phase noise modulation, which results in a power spectral density that is proportional to
the LO phase noise. This effect is called reciprocal mixing, and is illustrated in Fig. 5.37.
Phase noise is usually specified at a specific offset frequency from the LO carrier
frequency. For example phase noise of -100dBc/Hz4 at 1MHz offset means that the total
noise power in 1Hz bandwidth at 1MHz offset from the LO center frequency is 120dB
below the LO carrier power.
Interferer
signal
LO
f
After mixing
fRF0
Fig. 5.37 LO phase noise effect
The LO phase noise requirement is derived from the adjacent channel test. In the
case of Bluetooth, the derivation of LO phase noise is quite straightforward because the 4 LO phase noise is usually specified in dBc/Hz which represents the noise power in 1Hz bandwidth at the specified offset frequency relative to the LO fundamental output power.
146
interferer is narrow band Bluetooth modulated signal which we can approximate as a
single tone signal. If the adjacent channel power is Pint, at ∆f offset frequency, and the
signal power is Psig, then the signal to phase noise ratio (SPNR) is expressed as:
)2log(10/))2log(10( int BWPNICBWPNPPSPNR sig −−=++−= (5.21)
Where C/I is the carrier to interference ratio and BW is the channel selection filter
bandwidth. BW is multiplied by 2 to account for both sidebands around the offset
frequency. If the signal is x dB above sensitivity (x is specified in the standard), then the
signal to thermal noise ratio is xSNR +min (assuming same receiver gain distribution as in
the sensitivity test). Therefore, the overall SNR at the output of the receiver is given by:
10/)(10/10/
min1010110 xSNRSPNR
SNR+−− +
= (5.22)
To get the maximum LO phase noise, we set the overall SNR=SNRmin:
10/)(10/))2log(10/(10/
min
min
1010110 xSNRBWPNIC
SNR+−−−− +
= (5.23)
Therefore, the required LO phase noise, with some margin, is given by:
)101log(10)2log(10/ 10/min
xBWSNRICPN −−−−−= (5.24)
Table 5.5 shows the different interferer levels, offset frequencies and the
corresponding required LO phase noise for SNRmin = 13.6dB and BW = 600kHz.
Table 5.5 Required LO phase noise in Bluetooth mode
∆f (MHz) C/I (dB) x (dB) PN (dBc/Hz)
1 0 10 -74.85
2 -30 10 -104.85
3 -40 3 -117.41
147
At the offset frequencies listed in table 5.5, the LO phase noise is proportional to 1/f.
Therefore, the phase noise value at 3MHz offset is the toughest to achieve and hence will
be taken as the specification for the LO phase noise.
In Wi-Fi mode, the interferer must be considered as a wideband signal; therefore
equation (5.21) can be rewritten as:
)2log(10/
)10)(log(10 10/)(int
BWPNIC
dffCCKPPSPNR
eff
BW
BW
ffPNnsig
−−=
∗−−= ∫−
∆+
(5.25)
Where PNeff is the LO effective phase noise for a CCK signal, CCKn(f) is the
normalized (power = 1) CCK signal spectrum. Assuming 1/f dependence for phase noise,
therefore:
)2log(10)10log(10 10/)( BWdfff
KPNBW
BW
fCCKeff −
∆+∗= ∫−
(5.26)
In Wi-Fi, the interferer is at MHzf 25=∆ away at -35dBm, while the signal level is -
70dBm, 6dB above sensitivity. Substituting these values in equation (5.26), the required
effective phase noise is -111dBc/Hz. Simulations using MATLAB (appendix C) showed
that in order to obtain an effective PN of -111dBc/Hz for CCK, the actual phase noise at
25MHz is -99.2dBc/Hz, 13.8dB above the effective value. This is equivalent to -90dBm
at 3MHz offset if 1/f PN behavior is assumed, which is about 27.4dB more relaxed than
the required PN in Bluetooth mode. Since this assumed case is much more stringent that
the actual wideband adjacent channel, we conclude that the PN of the receiver will be
determined by the Bluetooth rather than the Wi-Fi standard.
148
5.4.3. System Design Verification
Once all the blocks specifications are determined, we need to make sure that the
overall system complies with the standards specifications by passing it through the
standards tests described in section 5.5.1.1. This is done by using the MATLAB code in
appendix D. In these tests, we have to make sure that the signal level at the input of
output of every block is within the dynamic range of that block and that the required
SNRmin is achieved at all input signal levels.
5.4.3.1. System verification in Bluetooth mode
Fig. 5.38 illustrates the signal level at the ADC input versus the RF input level. As
mentioned before, the LNA is activated and the VGA gain is set to minimum at startup.
In other words, the signal is initially assumed to be in the range from -57dBm to -33dBm
as indicated in Fig. 5.38. Based on the measured digital signal level at the output of the
ADC, a decision will be taken to bypass the LNA by the attenuator or to switch the VGA
gain to maximum or to keep the current gain distribution. The gain switching strategy is
depicted in Fig. 5.39 and is designed such that the signal level at the ADC input is below
4dBm and above the ADC quantization noise floor by at least 24dB. Fig. 5.40 shows the
signal and noise levels along the receiver chain in the Bluetooth sensitivity test (a target
sensitivity of -84dBm is assumed). The signal level at the ADC input is -23dBm, well
above the ADC quantization noise floor which is about -50dBm. The SNR at the input of
each block is depicted in Fig. 5.41. The adjacent channel test is shown in Fig. 5.42. Note
that in this test, the interferer level is at maximum (4dBm) at the filter input. The figure
shows that, as pointed out earlier, this test determines the maximum RF gain that can be
employed. The filter has 6dB gain and it attenuates the interferer by 54dB (signal gain –
149
interferer gain = 60dB). Fig. 5.43 depicts the Bluetooth maximum signal test. The
specified maximum Bluetooth is -20dBm. The LNA is bypassed by the -15dB attenuator
and the VGA gain is set to minimum (0dB). The signal level at the ADC input is -
13dBm, which means that even higher input RF signal powers can be tolerated in the
Bluetooth mode. At the input level -33dBm when the LNA/attenuator is switched, proper
operation of the receiver in either case (LNA is activated or bypassed) must be verified.
Fig. 5.44 shows the signal and noise levels when the LNA is active (high gain) and VGA
is at low (0dB) gain for the -33dBm signal level. When the LNA is bypassed by the
attenuator, the signal level at the ADC input drops as shown in Fig. 5.45. Figs. 5.40-5.45
are obtained from the MATLAB code in section D.1.
ADC input level
RF input level
4dBm
VGA maxLNA max
VGA minLNA max
VGA minLNA min
-26dBm
-22dBm
-33dBm
Rx gain = 61dB Rx gain = 7dBRx gain = 37dB
-57dBm -3dBm
-50dBm ADC noise floor
-84dBm
-23dBm
Fig. 5.38 ADC input level versus RF input power in Bluetooth mode
LNA max gainVGA min gain
Signal leve atADC input SADC
SADC > 4dBm
SADC< -22dBm
-22dBm< SADC<4dBm
switch LNA to min gain
switch VGA to max gain
keep current gain settings
Fig. 5.39 Gain switching strategy in Bluetooth mode
150
LNA Mixer Filter VGA ADC-120
-110
-100
-90
-80
-70
-60
-50
-40
-30
-20
Inp
ut
sig
nal
an
d n
ois
e le
vels
in
dB
m
Signal levelNoise level
LNA gain = HiVGA gain = Hi
Fig. 5.40 Signal and noise levels in Bluetooth sensitivity test
LNA M ixer Filter VGA ADC22
24
26
28
30
32
34
Inp
ut S
NR
leve
l in
dB
SNR
LNA gain = HiVGA gain = Hi
Fig. 5.41 Signal-to-noise ratio in Bluetooth sensitivity test
151
LNA Mixer Filter VGA ADC-70
-60
-50
-40
-30
-20
-10
0
10
Inp
ut
sig
nal
an
d i
nte
rfer
er le
vels
in
dB
m
Signal levelInterferer level
LNA gain = HiVGA gain = Hi
Fig. 5.42 Adjacent channel test in Bluetooth mode
LNA Mixer Filter VGA ADC-35
-30
-25
-20
-15
-10g
Sig
nal
leve
l in
dB
m
S ignal level
LNA gain = LoVGA gain = Lo
Fig. 5.43 Bluetooth maximum signal test
152
LNA Mixer Filter VGA ADC-120
-100
-80
-60
-40
-20
0
20
Inpu
t si
gn
al a
nd
no
ise
leve
ls in
dB
m S ignal levelNoise level
LNA gain = HiVGA gain = Lo
Fig. 5.44 Signal test at gain switching point (LNA gain is activated and VGA gain is low)
LNA Mixer Filter VGA ADC-120
-110
-100
-90
-80
-70
-60
-50
-40
-30
-20
Inpu
t si
gna
l an
d n
oise
leve
ls in
dB
m S igna l levelN o ise leve l
LNA gain = LoVGA gain = Lo
Fig. 5.45 Signal test at gain switching point (LNA is bypassed and VGA gain is low)
153
5.4.3.2. System verification in Wi-Fi mode
The same tests are performed in Wi-Fi mode. The signal level at the ADC input is
plotted versus the input RF signal level in Fig. 5.46. Note that due to the fine VGA gain
steps, the signal level at the ADC input is well controlled within ±1dBm for the entire
desired signal level range. The LNA is bypassed if the RF signal level is -37dBm. The
sensitivity test is performed at a target sensitivity of -80dBm. The gain switching strategy
is depicted in Fig. 5.47. Fig. 5.48 shows the signal and noise levels at the inputs of the
building blocks while the SNR is depicted in Fig. 5.49. The SNR at the ADC input is
about 19.8dB, about 6dB above SNRmin for Wi-Fi mode. This is consistent with the 6dB
margin taken in the NF calculations (equation (5.13)). The adjacent channel test is plotted
in Fig. 5.50. 50dB interferer attenuation by the channel selection LPF is assumed. Since
the interferer is 35dB higher than desired signal, the signal-to-interference ratio is 15dB
at the ADC input. The maximum signal test is depicted in Fig. 5.51. The RF input signal
level is -4dBm (for the low rate, 2Mbit/s, in 802.11 standard). LNA is bypassed by the
attenuator and the VGA gain is set at minimum. The signal level at the ADC input is
3dBm which is still within the ADC dynamic range. At the signal level -37dBm, the
receiver must operate properly whether the LNA or the attenuator is activated to insure
receiver compliance to the Wi-Fi standard at all signal levels. Fig. 5.52 shows the signal
and noise levels when the LNA is activated. The VGA gain is set to its minimum (0dB)
and the signal level at the ADC is at 0dBm. Fig. 5.53 depicts the test when the LNA is
bypassed by the attenuator at the same signal level (-37dBm). The SNR at the ADC input
is higher than 30dB and the signal level at the ADC input is 0dBm.
154
Therefore, all the plots ensure that, with the block specifications determined in the
previous subsection, the overall system is meeting the standards specifications. Fig. 5.48-
5.53 are obtained from the MATLAB code in section D.2.
ADC input level
RF input level
1dBm
VGA minLNA min
-7dBm
Rx gain = 7dB
-38dBmADC noise floor
-100dBm
-1dBm
-80dBm
Targetsensitivity
-37dBm
LNA minLNA maxVGA maxLNA max
Rx gain = 99dB
Fig. 5.46 ADC input level versus RF input power in Wi-Fi mode
LNA max gainVGA min gain
Signal leve atADC input SADC
SADC > 1dBm
SADC< -1dBm
switch LNA to min gain
keep LNA gain at max
-1dBm< SADC<1dBm keep current gain settings
find VGA gain so that-1dBm< SADC<1dBm
find VGA gain so that-1dBm< SADC<1dBm
Fig. 5.47 Gain switching strategy in Wi-Fi mode
155
LNA Mixer Filter VGA ADC-120
-100
-80
-60
-40
-20
0
Inp
ut
sig
nal
an
d n
ois
e le
vels
in d
Bm
S igna l leve lN oise level
Fig. 5.48 Signal and noise levels in Wi-Fi sensitivity test
LNA Mixer Filter VGA ADC19
20
21
22
23
24
25
26
27
Inpu
t S
NR
leve
l in
dB
S N R
Fig. 5.49 Signal-to-noise ratio in Wi-Fi sensitivity test
156
LNA Mixer Filter VGA ADC-70
-60
-50
-40
-30
-20
-10
0
Sig
nal
leve
l in
dB
m
S ignal levelInterferer level
Fig. 5.50 Adjacent channel test in Wi-Fi mode
LNA Mixer Filter VGA ADC-20
-15
-10
-5
0
5Maximum signal test in WiFi m ode
Sig
nal
lev
el in
dB
m
Signal leve l
Fig. 5.51 Wi-Fi maximum signal test
157
LNA Mixer Filter VGA ADC-120
-100
-80
-60
-40
-20
0
Inp
ut
sig
nal
an
d n
ois
e le
vels
in
dB
m
Signal levelNoise level
Fig. 5.52 Signal test at gain switching point in Wi-Fi mode (LNA gain is activated)
LNA Mixer Filter VGA ADC-120
-100
-80
-60
-40
-20
0
Inpu
t si
gnal
and
noi
se le
vels
in d
Bm
S igna l leve lN o ise level
Fig. 5.53 Signal test at gain switching point in Wi-Fi mode (LNA is bypassed)
158
CHAPTER VI
CHAMELEON: CIRCUIT IMPLEMENTATIONS
In this chapter, the implementation of every building block, from LNA to ADC, in
the dual mode Bluetooth/802.11b receiver will be presented. Each block was designed by
one or two PhD students. More design details will be given for the variable gain amplifier
(VGA) as it was my design assignment.
6.1. LNA and Mixer
The LNA circuit is shown in Fig. 6.1. It employs an inductively degenerated
differential pair structure. LNA input matching is established on chip using inductive
source degeneration technique [51]. In the presence of large signals, the LNA bias current
Itail is cut off and the signal is passed through a -15dB attenuator formed by MOS
transistors M5 – M9 in the triode region. A capacitor Cm is inserted into the attenuator to
improve impedance matching for the low gain mode. The only required off chip
component is a 1:1 balun for single-ended to differential conversion. In the high gain
mode, at small signal levels, all NMOS transistors of the attenuator M5 – M9 are turned
off by connecting their gates to ground. Thus, the normal operation of the LNA is not
affected. NMOS RF driving stage is used because MOS has better linearity than bipolar
for the same bias current. NPN cascode transistors are used rather than NMOS transistors
to have a more stable voltage at the drain of each of the NMOS RF driving transistors M1
and M2.
159
The mixer shown in Fig. 6.2 is a fully differential Gilbert-cell based structure. The I
and Q branches sharing the same RF drive stage, therefore eliminating the RF drive stage
mismatch present in the conventional two separate I/Q mixers. The current commutating
switches are NPN bipolar transistors which require less LO power than NMOS transistors
switch pairs. RF driving stage uses NMOS transistor for high linearity.
All of the biasing voltages and currents of the LNA and mixer are derived from a
PTAT current source to ensure temperature stability of the RF front-end.
Measurement results for the input impedance matching condition for the high and
low gain modes are shown in Figs. 6.3 and 6.4, respectively. In both cases, the input
reflection coefficient (S11) is less than -11dB. This means that, from the incident input RF
power to the LNA, only -11dB of that power is reflected back due to imperfect matching.
Summary of LNA/Mixer measurement results are listed in table 6.1.
LNA_cas_bias
Vo+
Vo-
VDD
LNA_bypass
Ls Ls
LdLd
LNA bypass switches and attenuator
LNA_rf_biasi_tail
CmVin+
Vin- M1 M2
M5
M6
M7
M8
M9
M3 M4
Q1 Q2
Rb Rb
Bond wire
Cd Cd
Vbb
Rb Rb
Fig. 6.1 LNA circuit
160
IFI
LO_bias
LOI
IFQ
LO_bias
LOQ
VDD
imix_bias
RF_in
Fig. 6.2 Mixer circuit
Fig. 6.3 Input matching for the high gain mode
161
Fig. 6.4 Input matching for the low gain mode
6.2. VCO and PLL
The LO signal is generated by an integer-N type frequency synthesizer covering the
specified band for both standards: from 2.4GHz to 2.5GHz in 1MHz steps. A typical
dead-zone free phase-frequency detector (PFD), with 1ns output pulses in lock condition,
is used. The PFD is followed by a charge pump (Fig. 6.5) with a cascode output [52]. The
cascode transistors provide a larger output resistance that reduces the output voltage
dependence of the output current. Switches Msp–Msn are sized to reduce the current
mismatch and switching time of the charge pump. The charge pump current Icp is 35µA.
A second order filter shown in Fig 6.6, uses a capacitance multiplier [53] to
implement capacitor C1 with a reduced area, minimum power consumption and negligible
noise contribution. Leakage current can be a problem in the capacitance multiplier if the
162
attenuation of the loop filter at the reference frequency is not large enough. A large
transistor length and small bias current are used to minimize the effect of the leakage
current on the reference spurs.
Vdd
Vdd
35 µA
UP
DWN
IoutVbiasN
VbiasP
Fig. 6.5 Cascode charge pump schematic
Vdd
C
5 µ
V
VbiasPIin
1:25
1:25
M1
M2
M4
M3
A
B
Fig. 6.6 Capacitance multiplier schematic
163
The VCO is operated at twice the LO frequency and the PLL is locked to a 2MHz
external reference crystal oscillator. A divide-by-two flip-flop generates a pair of
quadrature LO signals with a 1MHz tuning step. Even though more power is required to
operate the VCO and prescaler at twice the desired LO frequency, two important features
are improved under this strategy. First, the resultant phase and amplitude mismatch in the
quadrature output of the divide-by-two is very low: measured as 1° in phase and 2% in
amplitude. Second, the need for power hungry drivers for passive phase shifters is
avoided. Furthermore, to run the VCO at a different frequency from the power amplifier
of the transmitter is mandatory to avoid pull-in problems.
The VCO is implemented with a LC-tuned negative-gm oscillator as shown in Fig.
6.7. Selected BiCMOS technology provides some unique options for designing the
passive tank elements. Special low-resistance, top-metal layers is utilized for the on-chip
inductor. Simulated quality factor of the 1.5nH inductor is 13. Intrinsic base-collector
diode of bipolar device is used as a varactor, which provides ±17% capacitor variation
range. The varactor can provide the VCO with 760MHz of tuning range: wide enough to
overcome process and temperature variations. The inductor has a grid of deep trench
underneath that helps to isolate it from substrate-coupled interference. Several measures
have been taken to meet the phase noise requirement. Base nodes of bipolar transistor
drivers are AC-coupled with oscillating nodes and biased by an extra DC biasing circuit
to keep the transistors in the active region. Although the biasing circuit increases the
effective base resistance, improved linearity helps to reduce the overall phase noise.
Bypass capacitor on the common emitter node reduces noise the contribution of the
164
current bias transistors. The measured phase noise is -120dBc/Hz at a 3MHz offset and is
plotted in Fig. 6.8.
Vc
Vb
Ib
Fig. 6.7 VCO schematic
Fig. 6.8 Phase noise of the LO
165
A 15/16 dual-modulus phase switching prescaler follows the VCO [54]. The
prescaler is comprised of three stages of cascaded asynchronous dividers, a 8-to-1
multiplexer, phase selection circuitry and a final divide-by-two stage, as shown in Fig.
6.9a. The use of asynchronous frequency dividers reduces the power consumption of the
prescaler compared to conventional dividers operating at the same frequency, The output
of the third stage of dividers generate eight phases separated by a 45° each at a frequency
corresponding to 1/8 of the VCO frequency. When the modulus is set to 16 (Modulus
Control = ‘0’), no switching occurs and a single phase goes through the phase selection
block (Fig. 6.9b). When the modulus control signal is changed to 15 (Modulus Control =
‘1’), the phase selection block changes the output to the signal whose phase lags 45° the
current signal in every output cycle (i.e. p7 → p6). The phase switching scheme is
presented in Fig. 4b. Initially phase p7 is selected at the output of the multiplexer, when a
rising edge occurs in the output of the prescaler, the phase selector block changes the
output of the multiplexer to p6, which lags p7 by 45°. By changing the output of the
multiplexer from one phase to the one lagging it by 45°, the pulse width at the output of
the multiplexer is reduced, corresponding to skip the count one input cycle and increasing
the instantaneous division ratio.
166
fin /4fin
÷ 2IQIQ
f in /2
5GHz
2.5GHz 1.25
GHz
To Mixers LO Port
625MHz
÷ 2IQIQ
PhaseSelection
÷ 2
ModulusControl
÷ 2fout
IQIQ
÷ 2
PhaseSelection
fin /8
/2 /
2
/2
/2
/2
I
QI
Q
p0p2p4p6
p1p3p5p7
ModulusControl
Mux
a) Phase switching prescaler
1 2 3 4 5 6 7 1 2 3 4 5 6 7 8
1 2 3 4 1 2 3 4
1 2 3
fout
Mux
p7
p6
p5
b) Operation principle
Fig. 6.9 Phase switching prescaler
6.3. Channel Select Filter
The baseband 5th order LPF filter is implemented as an OTA-C structure with two
cascaded biquadratic sections and a single passive pole, as shown in Fig. 6.10. A
biquadratic section is shown in Fig. 6.11.
The proposed OTA architecture for the dual-mode filter is shown in Fig. 6.12. The
OTA used is a source degenerated transconductance, to enhance the limited linear input
range of the bipolar differential pair, with current scaling. The filter needs to provide a
167
programmable cut-off frequency ratio of 10:1. To avoid components values spread, both
resistors and capacitors are scaled by factors of 5 and 2, respectively.
Since the filter linearity requirement is almost the same in both modes, the bias
current is inversely scaled with the same factor as the resistance to keep the product
Ibias×R constant. This helps to optimize the power consumption in both modes. The active
loads M2 are also scaled accordingly by increasing the width of the load transistors so as
to keep the source-gate voltage constant in the two modes.
The first order passive pole used at the input stage of the filter helps improving the
out-of-band linearity. This improvement is due to the attenuation of the undesired
interferers. This is done at no cost in power. The cost is rather in area and noise. From
simulation results, adding the first order pole improves the out-of-band IIP3 by about
10dB and deteriorates the NF by about 1.5dB.
Lack of tracking between the single pole resistors and the transconductances of the
Gm cells, would result in pass-band ripple [55]. The transconductance of the source
degenerated OTA, shown in Fig. 6.12, is expressed as:
n
nRRg
RgR
Gm
mm +
×=+
×=1
11
1
1
1 where: Rgn m1= (6.1)
In order to minimize mismatches between the filter active poles and the first passive
pole, variations in the factor n/(n+1) in equation (6.1) must be minimized. Relative
variations in Gm can be expressed in terms of variations in n as:
∆
+=
∆nn
nGG
m
m
11 (6.2)
Therefore, in order to reduce the effect of process variations on the filter transfer
function, n should be increased. For example, if 101 == Rgn m , 50% process variations
168
in n corresponds to about 4.5% variations in Gm. It has been verified from simulations
that this ripple is less than 0.2dB which is acceptable for the current application.
Since the out-of-band blockers will be attenuated by the filter, harmonics generated
by the out-of-band blockers are dominated by the filter’s first stage. Hence, to improve
the overall filter linearity, the first stage is designed to have better linearity by using more
current Ibias in the first stage. Increasing the current of the first stage by 50% (the total
current by 10%) results in a 3dB improvement in IIP3 of the overall filter.
The dimensions of the switches in series with the degeneration resistors are chosen to
make the switch resistance about 1% of the resistor such that the linearity performance of
the OTA, and consequently the filter, is almost not affected.
The filter dissipates 2.7mA and 0.9mA for Wi-Fi and Bluetooth modes, respectively.
It has 6dB gain and occupies 0.9mm2 of silicon area. The measured frequency response
of the filter in both modes and for different tuning control bits is shown in Fig. 6.13.
Measured filter characteristics are summarized in table 6.1.
Biquad #1
fc=6MHzQ=0.62
Biquad #2
fc=6MHzQ=1.62
Vin+
Vin-
Vout+
Vout-
First Order LP
fc=6MHz
Fig. 6.10 Butterworth filter block diagram
169
_
++_
_
+
+
_
_
+
+
_
_
++_
Vin+
Vin-
VBP1+
VBP1-
VLP1-
VLP1+
CM
Info
rmat
ion
CM
Info
rmat
ion
gm3 gm2 gm1 gm1
C
C
C
C
+-
Vref+-
Vref
Vbias Vbias
CMFBcomparator
CMFBcomparator
Fig. 6.11 Biquadratic section
VCMFB
Vin+
VDD
Ibias
4M1
M2
Vout+Vout
-
VCMR
Q1
5R
R
Q1
5R
Vin-
M1M1
4M2 M2 4M2
VCMFB
VDD
C
C C
C
WiFi/BT
WiFi/BT WiFi/BT
WiFi/BT WiFi/BT
WiFi/BT
WiFi/BT
Fig. 6.12 Implementation of the dual-mode OTA
170
Fig. 6.13 Dual-mode operation and programmability of the LPF
Table 6.1 Summary of performance
Bluetooth mode 802.11b mode
Filter type Butterworth Butterworth
Filter order 5 5
Cut-off frequency 600kHz 6MHz
IIP3 40dBm 1 40dBm 2
Integrated input referred noise 90µVrms 100µVrms
Power supply 2.5V 2.5V
Current consumption 4 0.9mA 3 2.7mA 3
Area 4 0.9mm2 0.9mm2
1 Two tones applied at 1.83MHz and 3.16MHz 2 Two tones applied at 18.3MHz and 31.6MHz 3 Including CMFB and biasing circuitry 4 I and Q branches
171
6.4. Variable Gain Amplifier
6.4.1. DC Offset Problem, Revisited
If the receiver has no prior information regarding the incoming signal level, it has to
figure it out by itself when it receives the first bits of the incoming stream and then adapts
itself as fast as possible to receive and decode the information. The overall receiver gain
is adjusted, depending on the signal level, to keep almost constant signal level at the
ADC input through the VGA [56-60]. The most important factor that determines the
response time of the receiver is the settling time of its VGA, which is inversely
proportional to the VGA bandwidth. A number of wideband VGA designs can be found
in the literature [57, 60]. The problem becomes even more difficult for direct-conversion
receivers where the dc offset due to mismatches is a serious issue. In general, the output
dc offset of a VGA depends on its gain, and therefore, has to be corrected each time the
VGA gain is changed. Another requirement in the VGA is to have constant bandwidth
(and therefore, constant settling time) versus gain. Table 6.2 lists a summary of required
specifications for the VGA.
Table 6.2 Summary of required VGA specifications
Value Specification
Wi-Fi Bluetooth
Gain 0-62dB 0/24dB
Gain step 2dB 24dB
Bandwidth >10MHz >3MHz
Input referred noise < HznV /2.28 < HznV /3.50
1dB compression point >2Vpp >2Vpp
172
The BiCMOS process is used to save power consumption in the multi-standard
receiver. However, the VGA design has been done using only CMOS transistors which
means that it can also be implemented in a CMOS process.
Since the maximum gain of the VGA is 62dB in Wi-Fi mode, a 3-stage VGA is used
with about 20dB gain/stage. However, only the 3rd stage is used in Bluetooth mode to
provide 0/24dB gain. The VGA is used in a direct conversion receiver, and therefore, dc
offset generated by mismatches in the baseband blocks (including the VGA itself) may
saturate the stages of the VGA unless this offset is cancelled before it gets amplified by
the VGA stages. The offset can be cancelled between VGA stages by injecting a
correction signal from a digital-to-analog converter (DAC) as illustrated in Fig. 6.14 or
by using simple RC high-pass filters with low cutoff frequency as in Fig. 6.15. The cutoff
frequency must be chosen carefully: too low a cutoff might not remove enough 1/f noise,
whereas too high a cutoff may remove precious signal spectrum. According to BER
baseband simulations, 5kHz cutoff results in only 0.2dB degradation in the required
SNRmin in Wi-Fi mode. In Bluetooth mode, 1kHz cutoff results in 0.1dB SNRmin
degradation. The HPF solution will only work if the dc offset remains constant while the
signal is being received. However, since the gain of the VGA is adjusted during the
preamble period, the output offset of each VGA stage may change if its gain is altered.
This is because the capacitor in the HPF takes time (which is inversely proportional to the
HPF bandwidth) to charge/discharge to the new offset value.
173
VGA1+
DAC 1
Register 1
+
DAC 2
Register 2
+
DAC 3
Register 3
De-MUX
DACselection
ADCDC
offsetcontrol
Offset frommixer/filter
VGA2 VGA3
Fig. 6.14 DC offset correction using digital feedback
Stage 10,20dB
From LPF
Stage 2-4,-2,0,16,18,20dB
Stage 30,6,12,18,24dB
BT/WiFi
Reset
1kHz1kHz5kHz5kHz
VrefVref Vref Vref
Fig. 6.15 DC offset correction using passive HPF
6.4.2. Proposed Solution for HPF Slow Response
The proposed solution to the slow response of the HPF to dc offset variations is to
design the VGA stages such that its output referred offset is constant regardless of its
gain setting. To illustrate how to design such a VGA, consider first the OpAmp-R based
VGA shown Fig. 6.16(a). Single ended topology is shown for simplicity. Differential
architecture is used in all VGA stages. The OpAmp input referred offset is represented by
174
the voltage source VOS at the non-inverting terminal. The gain of the VGA stage is
controlled by the digital bit d1. The output of the VGA in Fig. 6.16(a) is given by:
OSinout VG
GdGV
GGdG
V )1(2
12111
2
12111 +++
+−= (6.3)
Note that the second term in the above equation, which represents the output referred
offset, depends on the digital control input d1. Therefore, the output dc offset of this VGA
circuit depends on its gain. To keep the output dc offset independent of gain, the VGA
stage shown in Fig. 6.16(b) is proposed. The output voltage of the modified VGA stage is
expressed as:
OSinout VG
GGV
GGdG
V )1(2
1211
2
12111 +++
+−= (6.4)
In this case, the resistor G12 is always included in the feedback loop and therefore,
the output referred offset is constant as indicated in equation (6.4). The gain is changed
by switching only the input terminal of the resistor G12 to the input or to ac ground. Note
that the output dc offset is constant only if the offset from the previous stage is
completely rejected. The circuit in Fig. 6.16(b) has constant feedback factor regardless of
its gain. This means that the bandwidth of the VGA will be independent of its gain. This
is a very important property in a VGA design. The main drawback in this circuit is its
finite input resistance which is also variable. Since each VGA stage is preceded by a
passive HPF that uses large resistance and capacitance values to achieve the required low
cut-off frequency, the output of this passive HPF has to be buffered before it's applied to
the VGA input. The buffer is implemented using a source follower with reduced output
resistance (to be discussed later) and therefore can generate its own offset that will be
multiplied by the variable gain of the VGA and produce a variable offset at the VGA
175
output. To circumvent this problem, a separate buffer is used for each resistor in the bank
of resistors at the VGA input and the gain is controlled by the switching the input of the
buffer, not the output as shown in Fig. 6.17. Therefore the buffer offset is always
included in the circuit and the VGA output offset is kept constant.
G2
G11
VOS
Vi Vo
G12G2
G11
VOS
Vi Vo
G12
d1 d1
(a) (b)
Fig. 6.16 (a) Conventional OpAmp-R VGA (b) Proposed constant output offset VGA
G2
G11Vi Vo
G12d
buffer
VOS
VOS2
VOS1
Fig. 6.17 Proposed constant output offset VGA circuit using OpAmp-R
6.4.3. OpAmp and Buffer Designs
To achieve a gain of about 20dB in a VGA stage, a 3-stage OpAmp architecture has
been chosen, as shown in Fig. 6.18 The input and output common mode voltages are set
to 1.5V. Simulations showed that the OpAmp dc gain is more than 70dB and the phase
176
margin is more than 60° at 20dB gain. The 3rd stage of the OpAmp is the buffer stage,
which drives only the feedback resistor G2 as shown in Fig. 6.17, and the resistors RCM
used to sense the common mode output of the OpAmp in the common mode feedback
(CMFB) circuit (Fig. 6.19). The compensation capacitor CC is used to stabilize the
differential mode (DM) and common mode (CM) loops. Although the OpAmp is always
used in a high DM gain configuration, the CM loop uses unity gain and its stability has to
be enhanced by adding a degeneration resistor RCMFB in the CMFB transconductor shown
in Fig. 6.19. It's also worth to emphasize the advantage of using the configuration in Fig.
6.16(b) that the feedback factor is always constant and therefore, the OpAmp will have
the same phase margin for all gain settings. This allows us to use fixed compensation
capacitor while keeping the OpAmp bandwidth constant for all VGA gains. The buffers
though, have to drive the input resistors of the VGA stage which are smaller or equal to
the feedback resistor. Since inter-stage HPF's are used, the input CM voltage of the
buffers doesn't have to be the same as the output CM voltage, and therefore simpler
buffers with dc level shift can be used. To improve linearity and gain accuracy, a buffer
with reduced output resistance, as shown in Fig. 6.20, is used. The low frequency output
resistance of the buffer is approximately given by:
32
31
mm
ddout gg
ggR
+= (6.5)
Where gmi and gdi are the small signal transconductance and output conductance of
the transistor Mi.
177
Vi+ Vi-
VDD
M1
M2 M3
M4 M5 M7
M9
M11
M13
M6
M8
M10
M12
Vo+ Vo-
Vcmfb
CCCC
Vbn
Vbp
Fig. 6.18 3-stage OpAmp schematic
VDD
M14
M16 M17
M18 M19
Vref
M15
Vcmfb
RCMFB=10k
Vbn
Vo+
Vo-
RCM=200k
RCM=200k
Fig. 6.19 CMFB circuit
178
M1 M2
CC1
M4
VDD
Vin
Vout
Vbufn
Vbufp
M3
Fig. 6.20 The buffer circuit
6.4.4. VGA Gain Lineup
The VGA gain has been distributed among the 3 stages as shown in Fig. 6.15. The
gain si distributed such that all the stages have almost the same maximum gain (about
20dB gain/stage). In the first stage, the gain can be either 0 or 20dB. Fine gain control is
achieved through the second and third stages. The third stage has 5 possible gain settings
from 0 to 24dB with 6dB gain steps. The second stage has 6 possible gain settings which
are -4, -2, 0, 16, 18, 20dB. This gain control distribution allows the overall VGA gain to
be controlled from -4 to 64dB with 2dB steps. The gain control is distributed such that
the number of gain steps/stage is reduced to avoid the design complexity in each stage.
6dB steps are used in the last stage to allow using R2R ladder network for gain control as
will be discussed later. The following sections describe the design of the VGA stages.
179
6.4.4.1. First stage
The first stage of the VGA has the most stringent requirements in terms of noise and
linearity. This is because the noise of this stage is multiplied by the gain of the other two
stages. On the other hand, the minimum gain for this stage is 0dB, which means that the
signal can be at full scale voltage (1Vp). The first stage VGA circuit is similar to the
circuit shown in Fig. 6.17 with Ω=== kGGG 50/19/12112 . Again, single-ended
topology is shown for simplicity. The gain of the first stage is 0dB when d1=0 and 20dB
when d1=1.
6.4.4.2. Second stage
The second stage of the VGA has more relaxed noise requirements than its first
stage. Therefore, in order to reduce the power consumption in the OpAmp, the input and
feedback resistors are scaled up to reduce the loading on the OpAmp. To have about the
same circuit bandwidth, all the transistors sizes as well as dc currents are scaled down by
the same factor. Therefore, the power consumption of the second stage is lower than that
of the first stage by the same factor. Since the resistors values in the first stage are already
high (50kΩ), the resistor values in the second stage cannot be increased much higher
because of area limitation. Fig. 6.21 shows the second stage VGA circuit. The gain of the
second stage is controlled with the digital inputs d2 to d5. Gain values and the
corresponding digital inputs are listed in the table 6.3.
The values of Ri1, Ri2, and Ri3 (where i=1,2) are chosen such that:
dBRRR
R
iii
i 4)log(20321
1 −=++
(6.6)
180
dBRRR
RR
iii
ii 2)log(20321
21 −=++
+ (6.7)
RF=500k
Vi
Vi
d3 R10=(1-α)RF
R11=1.585αRF
R12=6.121αRF
R13=4.862αRFd2
R20=(1-β)RF/9d5
R21=1.585βRF/9
R22=6.121βRF/9
R23=4.862βRF/9d4
d6
R1 = RF
R2 = RF/9
Fig. 6.21 Second stage VGA circuit
Table 6.3 Gain values of the second stage and the corresponding digital inputs
G2 in dB d2 d3 d4 d5 d6
-4 0 0 0 0 0
-2 0 1 0 0 0
0 1 1 0 0 0
16 0 0 0 0 1
18 0 1 0 1 1
20 1 1 1 1 1
181
The parameters α and β in Fig. 6.21 are used to scale the input resistance seen by the
buffers. Note that the equivalent resistance of the resistors Ri1 (i=0, … , 3) is equal to RF
and is independent of the value of α. Similarly, the equivalent resistance of the resistors
Ri2 (i=0, … , 3) is equal to RF/9 and is independent of β. The lower the values of α and β,
the lower the total resistance area, and the higher the required driving capability in the
buffers, and therefore the higher the current drain. This is an example of power-area
trade-off in this design. Note that in order to have positive resistance values, α and β are
positive numbers less than 1.
6.4.4.3. Third stage
The third stage of the VGA is designed to have 6dB gain steps from 0dB to 24dB.
Therefore, a resistive ladder network can be used at the input to realize the required
gains. Fig. 6.22 shows the third stage VGA circuit. Gain values and the corresponding
digital inputs are listed in the table 6.4.
RF=500k
Vi Vi
d72R1
R1
2R1
R1
2R1
R1
2R1
2R1
d8
d9
d10
d11
R1=RF/10
Fig. 6.22 Third stage VGA circuit
182
Table 6.4 Gain values of the third stage and the corresponding digital inputs
G3 in dB d7 d8 d9 d10 d11
0 0 0 0 1 0
6 0 0 1 0 0
12 0 1 0 0 0
18 1 0 0 0 0
24 1 1 1 1 1
6.4.5. Testing Results
Since there are 35 VGA gain settings in the -4 → 64dB gain range, 6 control bits are
needed to control the gain. A decoder should be built to decode the gain control bits into
the di (i=1, … , 11) bits required for the switches in the VGA stages. One bit control is
saved by using only the gain range 0 →62dB which is required for the Wi-Fi standard.
Therefore, only 5 control bits am (m=0, … , 4) are used in the design. Table 6.5 lists the
gain distribution over the three VGA stages for the 32 possible VGA gain settings. The
VGA has been integrated with the other blocks in the multi-standard BT/Wi-Fi receiver.
A single VGA branch consumes 1.05mA and occupies 0.5mm×0.7mm of silicon area.
Testing setup of the VGA is illustrated in Fig. 6.23. The measured input referred noise
density is HznV /26 at maximum gain. Fig. 6.24 shows the time domain response of
the VGA when an input signal of 40mV amplitude is applied. The gain of the VGA is
switched from 4 → 20 → 28 → 12 → 4dB using the digital control bits. Fig. 6.25 shows
the VGA settling time by zooming in the 20 → 28dB transition. The figure shows that the
VGA settling time is less than 200ns. Fig. 6.26 shows the frequency response of the VGA
for gain settings from 6 → 46dB with 10dB step. The VGA bandwidth is about 21MHz
183
and it is almost constant throughout the entire gain range. Measured input referred 1dB
compression point at 0dB gain is -0.5dBm. Summary of VGA experimental results is
listed in table 6.6.
Table 6.5 Gain distribution over the three VGA stages
Total Gain 1st stage gain (dB)
2nd stage gain (dB)
3rd stage gain (dB)
a4 a3 a2 a1 a0
0 0 0 0 0 0 0 0 02 0 -4 6 0 0 0 0 1 4 0 -2 6 0 0 0 1 0 6 0 0 6 0 0 0 1 1 8 0 -4 12 0 0 1 0 0 10 0 -2 12 0 0 1 0 1 12 0 0 12 0 0 1 1 0 14 0 -4 18 0 0 1 1 1 16 0 -2 18 0 1 0 0 0 18 0 0 18 0 1 0 0 1 20 0 -4 24 0 1 0 1 0 22 0 -2 24 0 1 0 1 1 24 0 0 24 0 1 1 0 0 26 0 20 6 0 1 1 0 1 28 0 16 12 0 1 1 1 0 30 0 18 12 0 1 1 1 1 32 0 20 12 1 0 0 0 0 34 0 16 18 1 0 0 0 1 36 0 18 18 1 0 0 1 0 38 0 20 18 1 0 0 1 1 40 0 16 24 1 0 1 0 0 42 0 18 24 1 0 1 0 1 44 0 20 24 1 0 1 1 0 46 20 20 6 1 0 1 1 1 48 20 16 12 1 1 0 0 0 50 20 18 12 1 1 0 0 1 52 20 20 12 1 1 0 1 0 54 20 16 18 1 1 0 1 1 56 20 18 18 1 1 1 0 0 58 20 20 18 1 1 1 0 1 60 20 16 24 1 1 1 1 0 62 20 18 24 1 1 1 1 1
184
VGA
Vin
VOCMVi+
Vi-
Vo+
Vo-
Vout
Oscilliscope
OPA3692
Gain control
5 bits
ModeselectionBT/WiFi
Arbitrary waveform generator
1 bit
THS4502
Fig. 6.23 VGA test setup
Fig. 6.24 Time domain response of the VGA
185
Fig. 6.25 Settling time of the VGA
Fig. 6.26 Frequency response of the VGA at different gain settings
186
Table 6.6 Summary of VGA experimental results
Parameter ValueGain 0-62dB
Gain step 2dB Bandwidth 21MHz
Current drain 1.05mA Input 1dB compression (0dB gain) -0.5dBm
Settling time 200ns Input referred noise (62dB gain) HznV /26
Area 0.35mm2 Power supply 2.5V
6.5. Analog to Digital Converter
As mentioned in the system level considerations, the ADC must provide a 62dB
dynamic range at 11MS/s, and a 48dB dynamic range at 44MS/s for the Bluetooth and
802.11b base-band signals, respectively. So far, the only reported ADC architecture for
multi-standard radios has been Σ∆ with adjustable OSR [61]. Due to the relatively wide
signal bandwidth of the Wi-Fi signal at 11Mbps, the use of a Σ∆ ADC in this receiver
would result in unacceptably high power consumption. On the other hand, a time-
interleaved Nyquist-rate ADC can provide a consistent dynamic range over a wide
frequency band while the overall ADC sampling rate can be configured by the number of
the interleaved ADC branches in operation. In addition, the dynamic range can be
controlled by the number of active stages if pipeline architecture is used. For this reasons,
time-interleaved pipeline architecture has been chosen to implement a low-power
configurable ADC for the Chameleon receiver.
The ADC architecture is shown in Fig. 6.27. It consists of a single front end sample-
and-hold and two time interleaved 11 bit pipeline ADCs with alternate 4 and 3 bit stages.
In the Wi-Fi mode, both pipeline branches operate at 22MHz to provide an overall
187
sampling frequency of 44MHz. The last stage in each pipeline branch can be disabled in
this mode to reduce power consumption while still providing the required dynamic range.
For the BT receiving mode, only one pipeline branch is activated and the sampling rate is
scaled down to 11MS/s. A multi-bit pipeline architecture has been chosen instead of the
traditional 1.5bit/stage design to reduce the number of MDACs, the implementation cost
of the lower bits stages and hence the power consumption. As an additional advantage,
the use of multi-bit stages significantly reduces the ADC delay, which is convenient for
the real-time operation of the receiver.
Several design measures were taken at the circuit-level design to optimize the power
consumption. OpAmp sharing technique, instead of two individual OpAmps, is applied in
the S&H. CMOS and Bipolar transistors are combined in the OpAmps of the S&H and
MDAC to achieve the requirements with minimum static bias current. The 3-bit sub-
ADC employs a flash architecture while the 4-bit ADC adopts 2X flash interpolation
structure to reduce the number of the preamps. The sampling circuit at the input of the
preamp is designed to avoid loading the MDAC of the previous stage, thus significantly
reducing the current consumed in capacitance charging. Bipolar transistors are employed
in the comparator to achieve the desired speed using minimum static current.
The digital correction in the pipeline and the multi-bit structure relax the sub-ADCs
specifications and gain error requirement. However, calibration is still needed to maintain
the 11-bit linearity in the first stage MDAC. Fig. 6.28 depicts the online digital
calibration scheme applied. The principle of the digital calibration has been introduced
before [62], but the calibration has to be performed offline, which is undesirable in a
receiver. In this design, an identical pipeline ADC branch is added to the existing parallel
188
ADCs. The first stage MDACs are calibrated one by one in rotation. The first stage of the
extra branch substitutes and functions as the stage in calibration. Thus, the calibration is
transparent to the circuit outside. All the first stage MDACs are calibrated using the same
lower bits stages in the extra pipeline branch, hence their performance becomes identical.
This helps to compensate the mismatch among the branches. System wise, the I/Q
mismatch can also be alleviated through this scheme. Since the calibration can be
conducted at a much lower frequency, once every msec in our case, the overhead in
power consumption is insignificant.
The measured SNR for the ADC is 60dB for the 5.5MHz 802.11b signal sampling at
44MS/s without disabling the last pipeline stage, and 64dB for the 550kHz BT signal at a
sampling rate of 11MS/s. Fig. 6.29 shows the measured SNR in both receiving modes
through a 14-bit DAC. The ADC power consumption are 20.2mW and 14.8mW for the
801.11b and BT receiving modes, respectively.
Dig
ital C
orre
ctio
n&
Mul
tiple
xer
SHAVin
4 bit 3 bit 4 bit 3 bit
4 bit 3 bit 4 bit 3 bit
Dig
ital C
orre
ctio
n&
Mul
tiple
xer
SHAVin
4 bit 3 bit 4 bit
4 bit 3 bit 4 bit
Dig
ital
Cor
rect
ion
SHAVin
4 bit 3 bit 4 bit 3 bit
BluetoothMode
802.11bMode
11-bit
9-bit
DACADC
-+ GMDACSub-ADC
Fig. 6.27 Time-interleaved pipeline ADC architecture
189
MSB
MSB LSB
MSB LSB
MSB REFERENCELSB
MSB LSB
Dig
ital C
alib
ratio
n
I
Q
LSB
0.5 0.6 0.7 0.8 0.9 1 1.1 1.2 1.3 1.4 1.50.5
0.6
0.7
0.8
0.9
1
1.1
1.2
1.3
1.4
1.5
Analog Input (V )
Dig
ital
Ou
tpu
t
RealIdeal
ERROR
Fig. 6.28 Online digital calibration
(a) (b)
Fig. 6.29 Measured SNR of the ADC in (a) Bluetooth mode (64dB at 11Msample/s) (b) WiFi mode (48dB at 44Msample/s)
190
CHAPTER VII
CHAMELEON: EXPERIMENTAL RESULTS
7.1. System Chip
The Chameleon dual-mode Bluetooth/Wi-Fi receiver was fabricated in IBM 0.25µm
BiCMOS technology. The total chip area including pads is 21mm2. The total active area
including pads and excluding unused space is 17.7mm2. Details about the areas of the
receiver blocks are in table 7.1. Note that the ADC is occupying about 58% of the
receiver active area. The die micrograph of the chip is shown in Fig. 7.1. The chip was
tested in a 4.5in×4.5in printed circuit board (PCB). The PCB was fabricated through PCB
EXPRESS. The PCB uses FR-4 material with thickness 0.031″ in order to have
reasonable track widths for impedance matching at 2.4GHz. Fig. 7.2 shows the simplified
block diagram of testing board. Photograph of the actual PCB is depicted in Fig. 7.3. It
measures 5.6mm×3.8mm.
Table 7.1 Receiver area breakdown
area (mm2) area (% )
LNA+Mixer 0.5 4.5
PLL 1.6 13.2
Filter 1.9 15.9
VGA 1.0 8.1
ADC 7.0 58.3
Total active area 12.1 100.0
pads area 5.6
active area+pads 17.7
unused space 3.3
Total active area 21.0
191
Fig. 7.1 Chameleon die micrograph
ADC
I Q
LNA
ADC
OPA3692
InstrumenationAmplifier
OPA3692
Matchingnetwork
Chemeleon chip
Balun
InstrumenationAmplifier
5bits
Gain control
Testing board
SMAconnector
4bits
LPF frequencytuning
Modeselection
Fig. 7.2 Simplified block diagram of the testing board for the Chameleon chip
192
Filter I&Qoutputs
VGA I&Qoutputs
Mixer I&Qoutputs
RF input
Fig. 7.3 Photograph of the PCB used to test the Chameleon chip
7.2. Experimental Results
7.2.1. BER vs. Input Power (Sensitivity)
Testing setup for sensitivity measurement is shown in Fig. 7.4. The sensitivity is
measured as follows: the receiver gain is set to maximum and the LNA input is
terminated by a 50Ω resistance. The integrated output noise ( oN ) of the VGA is
measured using the vector analyzer HP89410A. Then the signal is applied from the signal
generator Agilent E44320 to the LNA input. The generator output signal level is set such
that the signal level at the VGA output reaches minSNRNo + . At this point, the signal
193
generator output power is the sensitivity level measured at the output of the generator.
The corresponding signal level at the input of the LNA is given by:
nattenuatio dcable/boarpowergeneratorinputLNA atpower −= (7.1)
The cable/board attenuation is estimated by using a similar board, with the same
components values, and measuring the attenuation from the generator to the spectrum
analyzer.
In order to obtain the BER vs. input power curve, the generator output level is swept
from about 2 or 3 dBs below the sensitivity level to about 6dB above. For every input
level, the following steps are done:
(1) Adjust the receiver gain such that the signal at the VGA output is within the
ADC dynamic range.
(2) Measure the signal level at the VGA output
(3) The integrated noise at the VGA output measured by turning signal generator
off and terminating the LNA input with a 50Ω resistance.
(4) At the VGA output, calculate the output SNR = signal level – noise level.
(5) Obtain BER from the simulated BER-SNR curve. Take measured receiver
non-idealities into account.
Using the above technique, the BER vs. input power curves are obtained for
Bluetooth and Wi-Fi modes. Fig. 7.5 shows the receiver BER versus the input signal
power for Bluetooth mode and 802.11b mode at 11 and 5.5Mbps. The measured
sensitivities are –86dBm for 802.11b 11Mbps (for BER = 10-5) and –91dBm for
Bluetooth (for BER = 10-3).
194
ADC
I Q
LNA
ADC
OPA3692
InstrumenationAmplifier
OPA3692
Matchingnetwork
Chemeleon chip
Balun
InstrumenationAmplifier
5bits
Gain control
Testing board
SMAconnector
4bits
LPF frequencytuning
Modeselection
Fig. 7.4 Testing setup for NF measurement
-94 -93 -92 -91 -90 -89 -88 -87 -86 -85
10-4
10-3
10-2
Inpu t P ow er [dBm ]
Blu
etoo
th B
ER
10-5
10-6
10-5
10-4
10-7
802.
11b
(11M
bps,
5.5
Mbp
s) B
ER
802.11b, 11Mbps
802.11b, 5.5Mbps
Bluetooth
Fig. 7.5 BER performance of the dual-mode receiver
195
7.2.2. IIP3
Testing setup for IIP3 measurements is shown in Fig. 7.6. Two signal generators and
a power combiner are used to run the IIP3 two-tone test. The test is done by sweeping the
powers of two generators simultaneously and measuring the output fundamental and IM3
for every input power level. Since the two-tone test is usually done for out-of-band
signals, the fundamental tones are attenuated by the filter, affecting the IIP3 measurement
accuracy. Therefore, the fundamental power is measured when the input signal is in-
band.
To summarize, the IIP3 measurement is done by performing the following steps for
every generator power level:
• Set the two generators at the same power level.
• Set the frequency of the two generator frequencies, f1 and f2, according to the
specified two-tone test.
• Measure the IM3 component at 2f2 – f1 (f2 is closer to the LO than f1) at the VGA
output.
• Turn off the one of the generators and set the frequency of the other one to an in-
band frequency.
• Measure the fundamental component at the VGA output.
The above procedure is used to measure the IIP3 of the receiver. The receiver two-
tone IIP3 curve in Wi-Fi mode is plotted in Fig. 7.7. The tones are applied at 12 and
25MHz from the LO frequency, when the LNA is in the high gain mode. Measured
receiver IIP3 is –13dBm.
196
ADC
I Q
LNA
ADC
OPA3692
InstrumenationAmplifier
OPA3692
Matchingnetwork
Chemeleon chip
Balun
InstrumenationAmplifier
5bits
Gain control
Testing board
SMAconnector
4bits
LPF frequencytuning
Modeselection
Powercombiner
Fig. 7.6 Testing setup for IIP3/IIP2 two-tone test
-50 -45 -40 -35 -30 -25 -20 -15 -10-80
-60
-40
-20
0
20
40
60
2-Tone Input Pow er [dBm ]
Ou
tpu
t [d
Bm
]
I IP3
Fig. 7.7 Receiver IIP3 when LNA is in high gain mode (VGA gain = 12dB)
197
7.2.3. IIP2
The same connection setup shown in Fig. 7.6 is used for IIP2 two-tone test. The two
tones from the signal generators are applied at 12 and 13MHz away from the LO
frequency. The IIP2 is 10dBm and is plotted in Fig. 7.8. IIP2 tones are applied at 12 and
13MHz from the LO.
-60 -50 -40 -30 -20 -10 0 10 20
-60
-40
-20
0
20
40
60
80
2-Tone Input Pow er [dBm]
Ou
tpu
t [d
Bm
]
IIP2
Fig. 7.8 Receiver IIP2 when LNA is in high gain mode (VGA gain = 24dB)
Table 7.2 presents a performance summary for this design and the two most similar
previous reported implementations [37, 38]. Compared to previous implementations, our
receiver consumes less power, has better Bluetooth sensitivity and comparable Wi-Fi
sensitivity. Power consumption contributions of the receiver blocks is reported in Table
7.3. The temperature stability of the receiver performance is improved by a PTAT on-
198
chip bias of the RF-Front end, and the on-chip calibration capability of all base-band
blocks.
Table 7.2 Performance summary of the Chameleon receiver
BT WiFi BT WiFi BT WiFiReceiver Architecture Low-IF Direct Conversion Low-IF Direct Conversion Direct Conversion Direct Conversion
Offset cancellationChannel select filterBaseband amplifier
ADCFilter bandwidth 1MHz (BPF) 7.5MHz (LPF) 1MHz (BPF) 7.5MHz (LPF) 600kHz (LPF) 6MHz (LPF)
Sensitivity -82dbm -88dBm -80dBm -92dBm (0dB SNR) -91dBm -86dBmTechnology
Rx active current 46mA 65mA 41.3mA (w/ ADC) 45.6mA (w/ ADC)ADC active current - - - - 13.4mA 15.6mA
IIP3 -7dBm -8dBmIIP2 N/A N/A
Rx area (w/ pads)ADC area (w/ pads) - - - -
Supply voltage10mm2
2.7V 1.8V 2.5V
-13dBm20dBm 10dBm
N/A 16mm2 (transceiver) 9mm2 (w/o ADC)
-12dBm
0.35µm CMOS 0.18µm CMOS 0.25µm BiCMOS60mA
separate shared sharedNot included Not included Included
Programmable loop Injection at AGC input ac couplingseparate programmable programmable
[37] [38] This design
Table 7.3 Power consumption contribution of the receiver blocks
Wi-Fi BT Wi-Fi BTLNA/Mixer 29.2 32.9
PLL 26.9 30.3Filter 2.7 0.9 5.8 2.2VGA 2.1 0.9 4.5 2.2ADC 15.6 13.4 33.5 32.4Total 46.5 41.3 100.0 100.0
power consumption (in %)Block
13.612.5
Current drain in mA
199
CHAPTER VIII
CONCLUSIONS
The dissertation has examined in detail a number of issues related to wireless radio
design both at the system and at the building block level. At the system level, a low-IF
receiver design for Bluetooth standard has been presented. The receiver was implemented
in TSMC 0.35µm CMOS process. This receiver yields comparable or better performance
than most published commercial receivers that use more expensive advanced
technologies.
To target wider range of wireless applications, a dual-mode 802.11b/Bluetooth
receiver has been implemented in IBM 0.25µm BiCMOS process. Direct conversion
architecture was used for both standards to achieve maximum level of integration and
block sharing. Compared to recently reported dual-mode receivers, this receiver has a
lower current consumption in both operating modes, better Bluetooth sensitivity, and
comparable Wi-Fi sensitivity.
At the building block level, a pseudo differential OTA-C complex filter design for
the low-IF Bluetooth receiver has been implemented. A pseudo differential OTA is used
to comply with low voltage operation. To save area and power consumption, a sound
scheme for common-mode control is implemented using a minimum number of CMFF
and CMFB circuits. The filter uses a non-conventional frequency tuning circuit
architecture that has less silicon area and design complexity compared to the
conventional PLL.
200
To be used as frequency modulator in the Bluetooth transmitter, a ROM-less DDFS
has been implemented in AMI 0.5µm CMOS process. The DDFS architecture is based on
piecewise linear approximation of the sine function. The proposed architecture is shown
to have significant area and power savings at high clock rates compared to the recently
reported structures in the literature. This renders the proposed structure more suitable for
low power portable applications.
As part of the dual-mode receiver, variable gain amplifier with offset cancellation
technique has been implemented. Offset cancellation is done via low cutoff frequency
HPFs in between the VGA stages.
In summary, the contribution highlights of this dissertation are:
• 100MHz Direct digital frequency synthesizer.
• Complex filter with frequency tuning for low-IF Bluetooth receiver.
• Dual mode 802.11/Bluetooth receiver design.
• 0-62dB variable gain amplifier with offset cancellation scheme.
201
REFERENCES
[1] Specifications of the Bluetooth System, Version 1.0 B, Overland Park, KS, Bluetooth SIG, 1999.
[2] R. Morrow, Bluetooth Operation and Use. New York: McGraw-Hill, 2002.
[3] Specifications of the IEEE 802.11b System, New York: IEEE, 1999.
[4] J. Ross, The Book of Wi-Fi. San Francisco: No Starch Press, 2003.
[5] C. Andren, K. Halford, and M. Webster, "CCK, the New IEEE 802.11 Standard for 2.4GHz Wireless LAN," in International IC, Taipei, Taiwan, pp. 25-39, 2001.
[6] J. Lansford, A. Stephens, and R. Nevo, "Wi-Fi (802.11b) and Bluetooth Enabling Coexistence," IEEE Network, vol. 15, pp. 20-27, Sept.-Oct. 2001.
[7] G. F. Ross, "The Evolution of UWB Technology," in Radio and Wireless Conf., Boston, Massachusetts, pp. 1-4, Aug. 2003.
[8] "Revision of Part 15 of the Commission's Rules Regarding Ultra-wideband Transmission Systems,," in ET-Docket 98-153, FCC, http://www.fcc.gov/oet/ dockets/et98-153/, accessed on 10/26/2003.
[9] J. D. Taylor, Introduction to Ultra-Wideband Radar Systems. Boca Raton, FL: CRC Press, 1995.
[10] M. Rofheart, "XtremeSpectrum Multimedia WPAN PHY," in IEEE 802.15.3 Working Group Submission,, La Jolla, CA: IEEE, July 2000.
[11] B. Razavi, "Architectures and Circuits for RF CMOS receivers," in Proc. IEEE Custom Integrated Circuits Conf., pp. 393-400, May 1998.
[12] B. Razavi, "Design Considerations for Direct-Conversion Receivers," IEEE Trans. Circuits Syst. II, vol. 44, pp. 428-435, June 1997.
[13] J. Crols and M. S. J. Steyaert, "Low-IF Topologies for High-performance Analog Front Ends of Fully Integrated Receivers," IEEE Trans. Circuits Syst. II, vol. 45, pp. 269-283, Mar. 1998.
[14] K. S. Shanmugan, Digital and Analog Communication Systems. New York: Wiley, 1979.
[15] W. Sheng, B. Xia, A. E. Emira, C. Xin, A. Y. Valero-Lopez, S. T. Moon, and E. Sánchez-Sinencio, "A 3-V, 0.35µm CMOS Bluetooth Receiver IC," IEEE J. of Solid-State Circuits, vol. 38, pp. 30-42, Jan. 2003.
202
[16] B. Razavi, RF Microelectronics. Upper Saddle River, NJ: Prentice Hall, 1998.
[17] K. Philips, "A 4.4mW 76dB Complex Σ∆ ADC for Bluetooth Receivers," in IEEE Int. Solid State Circuits Conf., San Francisco, Feb. 2003.
[18] F. Behbahani, Y. Kishigami, J. Leete, and A. A. Abidi, "CMOS Mixers and Polyphase Filters for Large Image Rejection," IEEE J. of Solid-State Circuits, vol. 36, pp. 873-887, June 2001.
[19] B. Xia, C. Xin, W. Sheng, A. Y. Valero-lopez, and E. Sánchez-Sinencio, "A GFSK Demodulator for Low-IF Bluetooth Receiver," IEEE J. of Solid-State Circuits, vol. 38, pp. 1397-1400, Aug. 2003.
[20] P. Andreani, S. Mattisson, and B. Essink, "A CMOS gm-C Polyphase Filter with High Image Band Rejection," in 26th European Solid-State Circuits Conf., Stockholm, Sweden, Sep. 2000.
[21] J. Crols and M. Steyaert, "An Analog Integrated Polyphase Filter for a High Performance Low-IF Receiver," in Symp. VLSI Circuits Dig. Tech. Papers, Kyoto, Japan, pp. 87-88, June 1995.
[22] F. Behbahani, H. Firouzkouhi, R. Chokkalingam, S. Delshadpour, A. Kheirkhahi, M. Nariman, M. Conta, and S. Bhatia, "A Fully Integrated Low-IF CMOS GPS Radio with On-Chip Analog Image Rejection," IEEE J. Solid-State Circuits, vol. 37, pp. 1721-1727, Dec. 2002.
[23] P. T. M. van Zeijl, J.-W. Eikenbroek, P.-P. Vervoot, S. Setty, J. Tangenberg, G. Shipton, E. Kooistra, I. Keekstra, and D. Belot, "A Bluetooth Radio in 0.18 µm CMOS," in IEEE Int. Solid State Circuits Conf. (ISSCC) Dig. Tech. Papers, San Francisco, CA, pp. 86-87, Feb. 2002.
[24] A. S. Sedra, W. M. Snelgrove, and R. Allen, "Analogue Bandpass Filters Design by Linearly Shifting Real Low-Pass Prototypes," in Proc. IEEE Int. Symp. Circuits and Systems, Kyoto, Japan, pp. 1223-1226, 1985.
[25] S. A. Jantzi, K.W. Martin, and A. S. Sedra, "The Effects of Mismatch in Complex Bandpass Σ∆ Modulators," in Proc. IEEE Int. Symp. Circuits and Systems, pp 227-230, May 1996.
[26] B. Nauta and E. Seevinck, "Linear CMOS Transconductance Element for VHF Filters," Electronics Letters, vol. 25, pp. 448-450, Mar. 1989.
[27] D. A. Sunderland, R.A. Strauch, S.S. Wharfield, H.T. Peterson, and C. R. Cole, "CMOS/SOS Frequency Synthesizer LSI Circuit for Spread Spectrum Communications," IEEE J. of Solid-State Circuits, vol. 19, pp. 497-505, Aug. 1984.
203
[28] H. T. Nicholas, H. Samueli, and B. Kim, "The Optimization of Direct Digital Frequency Synthesizer Performance in the Presence of Finite Word Length Effects," in Proc. 42th Annual Frequency Control Symp., Baltimore, MD, pp. 357-363, June 1988.
[29] A. M. Sodagar and G. R. Lahiji, "Mapping from Phase to Sine-amplitude in Direct Digital Frequency Synthesizers Using Parabolic Approximation," IEEE Trans Circuits Syst., vol. 47, pp. 1452-1457, Dec. 2000.
[30] A. Bellaouar, M.S. O'brecht, A.M. Fahim, and M. I. Elmasry, "Low-power Direct Digital Frequency Synthesis for Wireless Communications," IEEE J. of Solid-State Circuits, vol. 35, pp. 385-390, Mar. 2000.
[31] S. Mortezapour and E. K. F. Lee, "Design of Low-power ROM-less Direct Digital Frequency Synthesizer Using Nonlinear Digital-to-Analog Converter," IEEE J. of Solid-State Circuits, vol. 34, pp. 1350-1359, Oct. 1999.
[32] J. Jiang and E. K. F. Lee, "A ROM-less Direct Digital Frequency Synthesizer Using Segmented Nonlinear Digital-to-Analog Converter," in IEEE Custom Integrated Circuits Conf., San Diego, CA, pp. 165-168, May 2001.
[33] A. N. Mohieldin, A. Emira, and E. Sánchez-Sinencio, "A 100MHz, 8mW ROM-less Quadrature Direct Digital Frequency Synthesizer," IEEE J. of Solid-State Circuits, vol. 37, pp. 1235-1243, Oct. 2002.
[34] C. Meenakarn and A. Thanachayanont, "100-MHZ CMOS Direct Digital Synthesizer with 10-bit DAC," in Proc. Asia-Pacific Conf. Circuits and Syst., Bangkok, Thailand, pp. 385-388, Oct. 2002.
[35] B. Provost and E. Sánchez-Sinencio, "On-chip Ramp Generators for Mixed-signal BIST and ADC Self-test," IEEE J. of Solid-State Circuits, vol. 38, pp. 263-273, Feb. 2003.
[36] D. A. Johns and K. Martin, Analog Integrated Circuit Design. New York: John Wiley & Sons, 1997.
[37] H. Darabi, J. Chiu, S. Khorram, H. Kim, Z. Zhou, E. Lin, S. Jiang, K. Evans, E. Chien, B. Ibrahim, E. Geronaga, L. Tran, and R. Rofougaran, "A Dual Mode 802.11b/Bluetooth Radio in 0.35µm CMOS," in IEEE Int. Solid State Circuits Conf. (ISSCC) Dig. Tech. Papers, San Francisco, CA, pp. 86-87, Feb. 2003.
[38] T. Cho, D. Kang, S. Dow, C. H. Heng, and B. S. Song, "A 2.4Ghz Dual-Mode 0.18µm CMOS Transceiver for Bluetooth and 802.11b," in IEEE Int. Solid State Circuits Conf. (ISSCC) Dig. Tech. Papers, San Francisco, CA, pp. 88-89, Feb. 2003.
[39] E. Chien, W. Feng, Y. A. Hsu, and L. Tse, "A 2.4GHz CMOS Transceiver and Baseband Processor Chipset for 802.11b Wireless LAN Application," in IEEE Int.
204
Solid State Circuits Conf. (ISSCC) Dig. Tech. Papers, San Francisco, CA, pp. 1-10, Feb. 2003.
[40] P. M. Stroet, R. Mohindra, S. Hahn, A. Schuur, and E. Riou, "A Zero-IF Single-chip Transceiver for Up to 22 Mb/s QPSK 802.11b Wireless LAN," in IEEE Int. Solid State Circuits Conf. (ISSCC) Dig. Tech. Papers, pp. 204-205, Feb. 2001.
[41] K. Y. Lee, S. W. Lee, Y. Koo, H. K. Huh, H. Y. Nam, J. W. Lee, J. Park, K. Lee, D. K. Jeong, and W. Kim, "Full-CMOS 2-GHz WCDMA Direct Conversion Transmitter and Receiver," IEEE J. of Solid-State Circuits, vol. 38, pp. 43-53, Jan. 2003.
[42] T. Arai and T. Itakura, "A Baseband Gain-controlled Amplifier with a Linear-in-dB Gain Range from 14dB to 76dB and a Fixed Corner Frequency DC Offset Canceler," in IEEE Int. Solid State Circuits Conf. (ISSCC) Dig. Tech. Papers, San Francisco, CA, pp. 136-137, Feb. 2003.
[43] A. Savla, A. Ravindran, J. Leonard, and M. Ismail, "System Analysis of a Multi-Standard Direct Conversion Wireless Receiver," in 45th Midwest Symp. on Circuits and Systems, Tulsa, OK, vol. 3, pp. 401-404, Aug. 2002.
[44] H. Darabi, S. Khorram, H.-M. Chien, M.-A. Pan, S. Wu, S. Moloudi, J. C. Leete, J. J. Rael, M. Syed, R. Lee, B. Ibrahim, M. Rofougaran, and A. Rofougaran, "A 2.4-GHz CMOS Transceiver for Bluetooth," IEEE J. Solid-State Circuits, vol. 36, pp. 2016-2024, Dec. 2001.
[45] R. Adler, "A Study of Locking Phenomena in Oscillators," Proc. IRE, vol. 34, pp. 351-357, June 1946.
[46] K. Kurokawa, "Injection Locking of Microwave Solid-State Oscillators," Proc. IEEE, vol. 61, pp. 1386-1410, Oct. 1973.
[47] P. T. M. v. Zeijl, J.-W. Eikenbroek, P.-P. Vervoot, S. Setty, J. Tangenberg, G. Shipton, E. Kooistra, I. Keekstra, and D. Belot, "A Bluetooth Radio in 0.18 µm CMOS," in IEEE Int. Solid State Circuits Conf. (ISSCC) Dig. Tech. Papers, San Francisco, CA, pp. 86-87, Feb. 2002.
[48] F. O. Eynde, J.-J. Schmit, V. Charlier, R. Alexandre, C. Sturman, K. Coffin, B. Mollekens, J. Craninckx, S. Terrijn, A. Monterastelli, S. Beerens, P. Goetschalckx, M. Ingels, D. Joos, S. Guncer, and A. Pontioglu, "A Fully-Integrated Single-chip SOC for Bluetooth," in IEEE Int. Solid State Circuits Conf. (ISSCC) Dig. Tech. Papers, pp. 196-197, Feb. 2001.
[49] M. Chen, K. H. Wang, D. Zhao, L. Dai, Z. Soe, and P. Rogers, "A CMOS Bluetooth Radio Transceiver Using a Sliding-IF Architecture," in IEEE Custom Integrated Circuits Conf., San Jose, CA, Sept. 2003.
205
[50] H. T. Friis, "Noise Figure of Radio Receivers," Proc. IRE, vol. 32, pp. 419-422, July 1944.
[51] D. K. Shaeffer and T. H. Lee, "A 1.5-V, 1.5-GHz CMOS Low Noise Amplifier," IEEE Journal of Solid-State Circuits, vol. 32, pp. 745-759, May 1997.
[52] W. Rhee, "Design of High-performance CMOS Charge Pumps in Phase-locked Loops," in IEEE International Symposium on Circuits and Systems, Orlando, FL, vol. 2, pp. 545-548, June 1999.
[53] S. Solis-Bustos, J. Silva-Martínez, F. Maloberti, and E. Sánchez-Sinencio, "A 60-dB Dynamic Range CMOS Sixth-order 2.4-Hz Low-pass Filter for Medical Applications," IEEE Trans Circuits Syst. II, vol. 47, pp. 1391-1398, Dec. 200.
[54] K. Shu, E. Sánchez-Sinencio, J. Silva-Martínez, and S. J. K. Embabi, "A 2.4-GHz Monolithic Fractional-N Frequency Synthesizer with Robust Phase-Switching Prescaler and Loop Capacitance Multiplier," IEEE J. of Solid-State Circuits, vol. 38, pp. 866-874, June 2003.
[55] T. Itakura, T. Ueno, H. Tanimoto, A. Yasuda, R. Fujimoto, T. Arai, and H. Kokatsu, "A 2.7-V, 200-kHz, 49-dBm, Stopband-IIP3, Low-noise, Fully Balanced gm-C Filter IC," IEEE J. Solid-State Circuits, vol. 34, pp. 1155-1159, Aug. 1999.
[56] A. Motamed, C. Hwang, and M. Ismail, "A Low-Voltage Low-Power Wide-Range CMOS Variable Gain Amplifier," IEEE Trans Circuits Syst., vol. 45, pp. 800-811, July 1998.
[57] M. Mostafa, H. Elwan, A. Bellaour, B. Kramer, and S. Embabi, "A 110MHz 70dB CMOS Variable Gain Amplifier," in IEEE International Symposium on Circuits and Systems, Orlando, FL, vol. 2, pp. 623-631, July 1999.
[58] H. Elwan, A. El-Adawi, M. Ismail, H. K. Olsson, and A. Soliman, "Digitally Controlled dB-Linear CMOS Variable Gain Amplifier," Electronics Letters, vol. 35, pp. 1725-1727, Sept. 1999.
[59] H. O. Elwan and M. Ismail, "Low Voltage Digitally Controlled dB-Linear CMOS VGA Circuit for Wireless Communication," in IEEE International Conference on Electronics, Circuits, and Systems, Lisboa, Portugal, pp. 341-344, Sept. 1998.
[60] J. J. F. Rijns, "CMOS Low-Distortion High-Frequency Variable-Gain Amplifier," IEEE J. of Solid-State Circuits, vol. 31, pp. 1029-1034, July 1996.
[61] G. Gomez and B. Haroun, "A 1.5 V 2.4/2.9 mW 79/50 dB DR Σ∆ Modulator for GSM/WCDMA in a 0.13µm Digital Process," in IEEE Int. Solid State Circuits Conf. (ISSCC) Dig. Tech. Papers, San Francisco, CA, pp. 306-307, Feb. 2002.
206
[62] S.-H. Lee and B.-S. Song, "Digital-domain Calibration of Multistep Analog-to-Digital Converters," IEEE Journal of Solid-State Circuits, vol. 27, pp. 1679-1688, Dec. 1992.
207
APPENDIX A
MATLAB CODES FOR CALCULATING THE OPTIMUM SLOPE
VALUES FOR THE DDFS
% This code calculates the optimum slope values for % specific phase resolution, number of segments, and % normalized resistance step global slope_res; global phase_res; global amp_res; global amp; global phase_step global seg_length; slope_res=8; phase_res=5; amp_res=8; amp=1; Nseg=8; seg_length=2^(phase_res+1)/Nseg; phase_step=pi/2/2^(phase_res+1); phase=linspace(0,pi/2,2^(phase_res+1)+1); ideal_sin=amp*sin(phase); offsety(1)=0; for i=1:Nseg ideal_curve=ideal_sin((i-1)*seg_length+1:i*seg_length+1)- ideal_sin((i-1)*seg_length+1); [segment(i,:) slope(i) yend(i)]= find_fixed_segment(offsety(i), ideal_curve); offsety(i+1)=yend(i)-ideal_curve(end); end real_curve(1:seg_length)=slope(1)*linspace(0.5,seg_length -0.5,seg_length)*phase_step; for j=2:Nseg for i=1:seg_length real_curve((j-1)*seg_length+i)=sum(slope(1:j-1))* seg_length*phase_step+slope(j)*(i-0.5)*phase_step; end end real_phase=linspace(0,pi/2-pi/2/2^(phase_res+1), 2^(phase_res+1))+pi/2/2/2^(phase_res+1);
208
% This function calculates the optimum slope in a % specific segment function [segment, slope, yend]=find_fixed_segment(offsety, curve) global slope_res; global phase_res; global amp_res; global amp; global phase_step global seg_length slope=2^slope_res/amp; segment=offsety+slope*(0:seg_length)*phase_step; max_error=max(abs(curve-segment)); while max_error>=max(abs(curve-segment)) max_error=max(abs(curve-segment)); slope=1/(1/slope+amp/2^slope_res); segment=offsety+slope*(0:seg_length)*phase_step; end yend=offsety+slope*seg_length*phase_step;
% This function calculates optimum slope of 1st segment function [segment, slope, yend]= find_first_segment(offsety, curve) global slope_res; global phase_res; global amp_res; global amp; global phase_step global seg_length slope=2^slope_res/amp; segment=slope*(0:seg_length)*phase_step; max_error=max(abs(curve-segment)); while max_error>=max(abs(curve-segment)) max_error=max(abs(curve-segment)); slope=1/(1/slope+amp/2^slope_res); segment=offsety+slope*(0:seg_length)*phase_step; endif slope ~=0 slope=slope-amp/2^slope_res; end yend=offsety+slope*seg_length*phase_step;
209
APPENDIX B
MATLAB CODES FOR GFSK DEMODULATOR
%%% This code is to calculate the BER for GFSK demodulator clear all; j=sqrt(-1); % define non-idealities samp=10; whp=1e3*2/samp/1e6; bwlp=600e3*2/samp/1e6; dcoff=0.1; deltaphi=5/180*pi; deltamag=0.1; Qbit=5; delay=9; EbN=13; Num_data=50000; % Generate random data data=2*round(rand(1, Num_data))-1; smpl_sp=Num_data*samp; sp_data=zeros(1, Num_data*samp); for i=1 : Num_data sp_data((i-1)*samp+1:i*samp)=data(i); end % Add Guassian pulse shape BT=.5; t=[-samp:1:samp]; ht=sqrt(pi*2/log(2))*BT/samp*exp(-(pi^2*2*(BT/samp)^2/log(2))*power(t,2)); psh_sig=conv(sp_data, ht); % BT/samp=B, the Guassin filter cause 1 bit delay % Define complex envelope of the GFSK signal ind=0.32; % norminal value, modulation index for BT is 0.28-0.35 phi=0; gfsk=zeros(1, smpl_sp-2*samp); for i=1 : smpl_sp-2*samp, phi=phi+psh_sig(i+2*samp); % Get rid of the first 2 bits to obtain % better synchronization gfsk(i)=cos(ind*pi*phi/samp+deltaphi/2)*(1+deltamag/2)+ j*sin(ind*pi*phi/samp-deltaphi/2)*(1-deltamag/2); end
210
% Match filters d00=ones(1, 4*samp); d01=[ones(1, 3*samp) (-1)*ones(1,samp)]; d10=[ones(1, 3*samp)*(-1) ones(1,samp)]; d11=ones(1, 4*samp)*(-1); for i=1:samp, d00(i+samp)=-1; d00(i+samp*2)=-1; d01(i+samp)=-1; d01(i+samp*2)=1; d10(i+samp)=1; d10(i+samp*2)=-1; d11(i+samp)=1; d11(i+samp*2)=1; end sd00=conv(d00, ht); sd01=conv(d01, ht); sd10=conv(d10, ht); sd11=conv(d11, ht); s00=sd00(2*samp+1: samp*4); s01=sd01(2*samp+1: samp*4); s10=sd10(2*samp+1: samp*4); s11=sd11(2*samp+1: samp*4); phi00=0; phi01=0; phi10=0; phi11=0; for i=1 : 2*samp, phi00=phi00+s00(i); mf00(i)=((exp(j*ind*pi*phi00/samp))').'; phi01=phi01+s01(i); mf01(i)=((exp(j*ind*pi*phi01/samp))').'; phi10=phi10+s10(i); mf10(i)=((exp(j*ind*pi*phi10/samp))').'; phi11=phi11+s11(i); mf11(i)=((exp(j*ind*pi*phi11/samp))').'; end d000=[ones(1, samp) -ones(1, 3*samp) ones(1, samp)]; d001=[ones(1, samp) -ones(1, 2*samp) ones(1, samp) -ones(1, samp)]; d010=[ones(1, samp) -ones(1, samp) ones(1, samp) -ones(1, samp) ones(1, samp)]; d011=[ones(1, samp) -ones(1, samp) ones(1, 2*samp) ]; d100=[-ones(1, samp) ones(1, samp) -ones(1, samp) -ones(1, samp) ones(1, samp)]; d101=[-ones(1, samp) ones(1, samp) -ones(1, samp) ones(1, samp) -ones(1, samp)]; d110=[-ones(1, samp) ones(1, 2*samp) -ones(1, samp) ones(1, samp)]; d111=[-ones(1, samp) ones(1, 3*samp) -ones(1, samp)];
211
sd000=conv(d000, ht); sd001=conv(d001, ht); sd010=conv(d010, ht); sd011=conv(d011, ht); sd100=conv(d100, ht); sd101=conv(d101, ht); sd110=conv(d110, ht); sd111=conv(d111, ht); s000=sd000(2*samp+1: samp*5); s001=sd001(2*samp+1: samp*5); s010=sd010(2*samp+1: samp*5); s011=sd011(2*samp+1: samp*5); s100=sd100(2*samp+1: samp*5); s101=sd101(2*samp+1: samp*5); s110=sd110(2*samp+1: samp*5); s111=sd111(2*samp+1: samp*5); phi000=0; phi001=0; phi010=0; phi011=0; phi100=0; phi101=0; phi110=0; phi111=0; for i=1 : 3*samp, phi000=phi000+s000(i); mf000(i)=((exp(j*ind*pi*phi000/samp))').'; phi001=phi001+s001(i); mf001(i)=((exp(j*ind*pi*phi001/samp))').'; phi010=phi010+s010(i); mf010(i)=((exp(j*ind*pi*phi010/samp))').'; phi011=phi101+s011(i); mf011(i)=((exp(j*ind*pi*phi011/samp))').'; phi100=phi100+s100(i); mf100(i)=((exp(j*ind*pi*phi100/samp))').'; phi101=phi101+s101(i); mf101(i)=((exp(j*ind*pi*phi101/samp))').'; phi110=phi110+s110(i); mf110(i)=((exp(j*ind*pi*phi110/samp))').'; phi111=phi111+s111(i); mf111(i)=((exp(j*ind*pi*phi111/samp))').'; end
212
% Add white noise, define Eb/N %EbN=(8:0.5:16); for k=1:length(EbN), sigma=sqrt(samp/2/power(10,(EbN(k)/10))); %sigma=0; Nn=length(gfsk); No_I=randn(1, Nn)*sigma; No_Q=randn(1, Nn)*sigma; % LPF filters the noise order=5; [BBS,ABS]=butter(order, bwlp); % order=4; % [BBS,ABS]=cheby1(order, 0.5, bwlp); % Filter the noise through LPF Nk_I=filter(BBS,ABS,No_I); Nk_Q=filter(BBS,ABS,No_Q); gfsk0=filter(BBS,ABS,gfsk); % gfsk0=gfsk; % Let the signal, noise go through the HPF twice [BHP,AHP]=butter(1,whp, 'high'); Nk_I1=filter(BHP,AHP,Nk_I); Nk_Q1=filter(BHP,AHP,Nk_Q); gfsk1=filter(BHP,AHP,gfsk0); Nk_I2=filter(BHP,AHP,Nk_I1); Nk_Q2=filter(BHP,AHP,Nk_Q1); gfsk2=filter(BHP,AHP,gfsk1); % Received signal rx_sigo=gfsk2+(Nk_I+j*Nk_Q)+dcoff; rx_sig=[rx_sigo(delay+1:end) zeros(1,delay)]; % Quantization smpl_rx=length(rx_sig); Nrx=smpl_rx/samp; Ts=linspace(0, 1e-6*Nrx, smpl_rx)'; RxReal=(real(rx_sig)).'; RxImg=(imag(rx_sig)).'; qStep=2^Qbit; qinv=4/qStep; DataIn=[RxReal RxImg];
213
%Let the received signal go through the quantizer SimStp=Ts(2)-Ts(1); options=simset('Reltol',1e-3,'Maxstep',SimStp); sim('quantizer', Ts(end), options); rx_sig=Iout.'+j*(Qout.'); % Run the received signal through match filter bank % Calculate the first bit lambda(1)=abs(sum(rx_sig(1:2*samp).*mf00)); lambda(2)=abs(sum(rx_sig(1:2*samp).*mf01)); lambda(3)=abs(sum(rx_sig(1:2*samp).*mf10)); lambda(4)=abs(sum(rx_sig(1:2*samp).*mf11)); [mxl, ix]=max(lambda); Dr(1)=floor(ix/3)*2-1; for i=2:(length(rx_sig)/samp-1) icur=samp*(i-2); pSt=icur+1; pEd=icur+samp*3; if Dr(i-1)==1 lambda(1)=abs(sum(rx_sig(pSt:pEd).*mf100)); lambda(2)=abs(sum(rx_sig(pSt:pEd).*mf101)); lambda(3)=abs(sum(rx_sig(pSt:pEd).*mf110)); lambda(4)=abs(sum(rx_sig(pSt:pEd).*mf111)); else lambda(1)=abs(sum(rx_sig(pSt:pEd).*mf000)); lambda(2)=abs(sum(rx_sig(pSt:pEd).*mf001)); lambda(3)=abs(sum(rx_sig(pSt:pEd).*mf010)); lambda(4)=abs(sum(rx_sig(pSt:pEd).*mf011)); end [mxl, ix]=max(lambda); Dr(i)=floor(ix/3)*2-1; end Dr=Dr(9:(end-8)); Ds=data(10:Num_data-10); De=Dr-Ds; err=find(De); BER(k)=(length(err))/length(Ds)*100; [np f]=psd(Nk_I, 8192, samp*1e6); [sp f]=psd(real(gfsk2), 8192, samp*1e6); snr(k)=10*log10(sum(sp)/sum(np)); end BER
214
APPENDIX C
MATLAB CODES FOR CALCULATING EQUIVALENT IIP3, IIP2,
AND LO PHASE NOISE IN WI-FI
%%% This code is to calculate the equivalent IIP3 for CCK %%% interference in Wi-Fi standard clear all CCK_BB_spectrum % load CCK baseband spectrum with 3 side % lobes (44MHz wide) CCK=[CCK(end:-1:1)' CCK']'; % shift CCK spectrum to 44MHz CCK=CCK/max(abs(CCK)); % normalize CCK spectrum %3rd order harmonic HD3=conv(conv(CCK,CCK),CCK); % HD3 spectrum figure(1) hold off freq=linspace(0,3*4*22e6,length(HD3)); % frequency span of HD3 semilogy(freq,HD3) % HD3 spectrum plot HD3_center=length(HD3)/2; CCK_lobe=length(CCK)/8; % width of CCK side lobe % finding the edges of the band of the HD3 spectrum that passes % through the LPF left_edge=fix(HD3_center-(25+6)/11*CCK_lobe); right_edge=fix(HD3_center-(25-6)/11*CCK_lobe); Wn1 = 2*pi*6e6; % LPF bandwidth [B1 A1] = butter(5,Wn1,'s'); % filter centered at 25MHz from the HD3 center W1=linspace(2*pi*(25-3*4*11)*1e6,2*pi*(25+3*4*11)*1e6,length(HD3)); H1=abs(freqs(B1, A1, W1)); % filter frequency response HD3_inband=H1'.*HD3; % filtered HD3 hold on semilogy(freq,HD3_inband,'r') % plot filtered HD3 Pin=sum(CCK.^2)*4*22e6/length(CCK); % calculate CCK power % generate two tones with the same power as CCK % such that HD3 falls inband twotone=zeros(1,length(CCK));
215
%%% This code is to calculate the equivalent IIP2 for CCK %%% interference in Wi-Fi standard clear all CCK_BB_spectrum % load CCK baseband spectrum with 3 side lobes % 44MHz wide) CCK=[CCK(end:-1:1)' CCK']'; % shift CCK spectrum to 44MHz CCK=CCK/max(abs(CCK)); % normalize CCK spectrum %3rd order harmonic HD3=conv(conv(CCK,CCK),CCK); % HD3 spectrum figure(1) hold off freq=linspace(0,3*4*22e6,length(HD3)); % frequency span of HD3 semilogy(freq,HD3) % HD3 spectrum plot HD3_center=length(HD3)/2; CCK_lobe=length(CCK)/8; % width of CCK side lobe % finding the edges of the band of the HD3 spectrum that passes % through the LPF left_edge=fix(HD3_center-(25+6)/11*CCK_lobe); right_edge=fix(HD3_center-(25-6)/11*CCK_lobe); Wn1 = 2*pi*6e6; % LPF bandwidth [B1 A1] = butter(5,Wn1,'s'); % filter centered at 25MHz from the HD3 center W1=linspace(2*pi*(25-3*4*11)*1e6,2*pi*(25+3*4*11)*1e6,length(HD3)); H1=abs(freqs(B1, A1, W1)); % filter frequency response HD3_inband=H1'.*HD3; % filtered HD3 hold on semilogy(freq,HD3_inband,'r') % plot filtered HD3 Pin=sum(CCK.^2)*4*22e6/length(CCK); % calculate CCK power % generate two tones with the same power as CCK % such that HD3 falls inband twotone=zeros(1,length(CCK)); twotone(fix((44-1*25/3)/11*CCK_lobe))=sqrt(Pin);
216
%%% This code is to calculate the equivalent LO phase noise for CCK %%% interference in Wi-Fi standard clear all CCK_BB_spectrum % load CCK baseband spectrum with 3 side lobes (44MHz % wide) CCK=[CCK(end:-1:1)' CCK']'; % shift CCK spectrum to 44MHz CCK=CCK/max(abs(CCK)); % normalize CCK spectrum CCK_center=length(CCK)/2; CCK_lobe=length(CCK)/8; % width of CCK side lobe % shift CCK spectrum to 25MHz CCK=CCK(CCK_center-25/11*CCK_lobe:CCK_center+25/11*CCK_lobe); semilogy(CCK) % Generate a single tone at 25MHz single_tone=zeros(size(CCK)); single_tone(end/2)=sqrt(sum(CCK.^2)*2*25e6/length(CCK)); single_tone=single_tone+1e-10; % add eps to the tone to plot it in log scale figure(1) hold off freq=linspace(0,2*25e6,length(CCK)); % Frequency span of CCK spectrum=25MHzPN25MHz=-99.2124; % PN at 25MHz in dBc/Hz K=25e6*10^(PN25MHz/10); %PN constant PN=sqrt(K./freq); % Phase noise spectrum PN(1)=1*PN(2); % avoid dividing by zero semilogy(freq,CCK) % plot CCK spectrum hold on semilogy(freq,PN) % plot PN semilogy(freq,single_tone,'-') % plot the single tone No_CCK=conv(PN(end:-1:1), CCK); % Noise due to the CCK interferer No_single_tone=conv(PN(end:-1:1), single_tone); % noise due to the single tone figure(2) freq2=linspace(-2*25e6,2*25e6,length(No_CCK)); % frequency span of the noise hold off semilogy(freq2,No_CCK) % plot noise due to CCK interferer hold on semilogy(freq2,No_single_tone) % plot noise due to single tone interferer % find the noise band edges passed through the LPF N=length(No_CCK); left_edge=N/2-6/100*N; right_edge=N/2+6/100*N; % calculate equivalent PN for the CCK interferer PNeff = 10*log10(sum(No_CCK(left_edge:right_edge).^2)/sum(CCK.^2))-10*log10(2*6e6)
217
APPENDIX D
MATLAB CODES FOR SYSTEM LEVEL CALCULATIONS
D.1. Bluetooth System Design MATLAB Code
%This file is for Bluetooth receiver system level design% %Author: Ahmed Emira% %Below is the flow chart of the system level design % _________________ __________________ % | Baseband | | Standard | % | Requirements | | Specifications | % | (SNRmin, C/Imin) | |_________________| % |_________________| / % \ / % \ / % \ / % \ / % _____________________ % | Derive required | % | system level specs | % |____________________| % | % __________\|/_________ % | Blocks’ specifications | _______ % |____________________| | % | | % __________\|/_________ | % | Calculate system specs | /|\ % |____________________| | % | | % ___________\|/___________ | NO % | Do we meet required specs |_____ | % |_______________________ | % | YES % _____\|/_____ % | Stop | % |___________ |
218
% All IIP3's are in dBm referenced to 50Ohm, unless otherwise specified % All NF's are in dB referenced to 50Ohm, unless otherwise specified % All gains are in dB unless otherwise specified %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%% Constants %%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% K = 1.38e-23; % Boltzmann constant T = 300; % room temperature %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%% Baseband Requirements %%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% SNR=13; % dB C2I=13; % minimum Carrier to inband interference ratio (dB) %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%% Standard specifications %%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% Sensitivity = -84; % dBm, 15dB better than standard max_signal = -20; % dBm Rb = 1e6; %Mbps BW=0.6e6; %MHz Eb_No=SNR-10*log10(Rb/BW); % Adjacentchannel specs Signal_level_C2I=-70+3; %dBm C2I_3MHz = -40; %dBm Interferer_level_C2I= Signal_level_C2I-C2I_3MHz; Signal_level_IM3=-70+6; %dBm Interferer_level_IM3 = -39; %dBm C2I_IM3 = Signal_level_IM3-Interferer_level_IM3; %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%% General parameters %%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% Signal_ADC_max = 4; % this is the maximum signal level at ADC input(dBm) %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%% Required system specifications %%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% NF_margin = 6; Req_syst_NF = Sensitivity - (-174+10*log10(Rb)) - Eb_No - NF_margin IIP3_margin = 3; IM3 = Signal_level_IM3-C2I; Req_syst_IIP3 = (3*Interferer_level_IM3-IM3)/2+IIP3_margin Req_syst_IIP2 = 2*Interferer_level_C2I-Signal_level_C2I+C2I-6 % 6dB is subtracted since it is a single tone test % No IP2 test is specified Rx_Gain_max = Signal_ADC_max-Sensitivity; Rx_Gain_min = Signal_ADC_max-max_signal;
219
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%% Block specifications %%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % indices [RF_filter = 1, LNA = 2, Mixer = 3, Filter = 4, VGA = 5] NB=6; %number of blocks RF_filer = 1; % indices LNA = 2; Mixer = 3; Filter = 4; VGA = 5; ADC = 6; ADC_SNR = 60; % in dB ADC_NF = Signal_ADC_max - ADC_SNR - 10*log10(4*K*T*BW/2*1000); % RF_filter LNA Mixer Filter VGA ADC IIP2 = [ 100 10 47 63 30 40]; % in dBm IIP3 = [ 100 -8 5 23 10 10]; % in dBm NF_Gmax = [ 0 3 20 36 30 ADC_NF]; % in dB at max gain NF_Gmin = [ 0 15 20 36 40 ADC_NF]; % in dB at min gainGain_max = [ 0 15 16 6 24 0 ]; % max gain in dB Gain_min = [ 0 -15 16 6 0 0 ]; % min gain in dB Atten_3M = [ 0 0 0 60 0 0]; % interference % attenuation % relative to the % max gain Mixer_feedthrough = -30; %RF to IF isolation in dB %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%% LO phase noise %%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% LO_PN = -Interferer_level_C2I-Eb_No -10*log10(Rb) +10*log10(10^(Signal_level_C2I/10) - 10^(Sensitivity/10))-NF_margin % Phase noise in dBc/Hz at 25MHz %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%% System specifications %%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% for i=0:NB Gain_max_accum(i+1) = sum(Gain_max(1:i)); % accumulated gain from antenna end % to block input for i=0:NB-1 Int_atten_accum(i+1) = sum(Atten_3M(1:i)); % accumulated gain from antenna end % to block input for the interferer IIP3_eff = IIP3+1.5*Int_atten_accum; % effective IP3 IIP2_eff = IIP2+2*Int_atten_accum; % effective IP2 IIP2_eff(LNA) = IIP2_eff(LNA) - (Mixer_feedthrough - Gain_max(Mixer)); % IM2 % of the LNA is attenuated by the mixer's feedthrough
220
IIP2_IR_Lin = 10.^((IIP2_eff-Gain_max_accum(1:NB))/20); % IIP2 referred to the % Rx input in SQRT(mW) IIP3_IR_Lin = 10.^((IIP3_eff-Gain_max_accum(1:NB))/10); % IIP3 referred to the % Rx input in mW NF_IR_Lin = (10.^(NF_Gmax/10)-1)./10.^(Gain_max_accum(1:NB)/10)+1; % NF % referred to % the Rx input in % linear units Syst_NF = 10*log10(sum(NF_IR_Lin-1)+1) % System NF in dB Syst_IIP3 = 10*log10(1/sum(1./IIP3_IR_Lin)) % System IIP3 in dBm Syst_IIP2 = 20*log10(1/sum(1./IIP2_IR_Lin)) % System IIP2 in dBm %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%% IIP2, IIP3, and Noise contributions of different blocks %%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% IIP2_contributions = 10^(Syst_IIP2/20)./IIP2_IR_Lin(2:end); figure LABELS_IIP2 = 'LNA','Mixer','Filter','VGA'; pie(IIP2_contributions, LABELS_IIP2); title('IIP2 contributions of different blocks') NF_contributions = (NF_IR_Lin-1)/(10.^(Syst_NF/10) -1); LABELS_NF = 'RF Filter','LNA','Mixer','Filter','VGA'; figure pie(NF_contributions, LABELS_NF); title('Noise contributions of different blocks') IIP3_contributions = 10.^(Syst_IIP3/10)./IIP3_IR_Lin(2:end); figure LABELS_IIP3 = 'LNA','Mixer','Filter','VGA'; pie(IIP3_contributions, LABELS_IIP3); title('IIP3 contributions of different blocks') %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%% Test scenarios %%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%% Sensitivity test %%%% Sig_level_IR = Sensitivity; % signal level at antenna (Rx input referred) Noise_level_IR = -174+10*log10(BW); % thermal noise at antenna Sig_level(1:NB) = Sig_level_IR+Gain_max_accum(1:NB); % signal level at each % block input for i=0:NB-1 NF_IR_accum_Lin(i+1)=sum(NF_IR_Lin(1:i)-1)+1; end
221
Noise_level(1:NB) = 10*log10(10.^((Noise_level_IR+Gain_max_accum(1:NB))/10).*NF_IR_accum_Lin);SNR = Sig_level-Noise_level; figure plot([0:NB-2],Sig_level(2:end), 'k-o',[0:NB-2], Noise_level(2:end), 'k-s') set(gca,'XTick',[0:NB]) set(gca,'XTickLabel','LNA';'Mixer';'Filter';'VGA';'ADC') grid on title('Sensitivity test') ylabel('Input signal and noise levels in dBm') legend('Signal level','Noise level',0) figure plot([0:NB-2], SNR(2:end), 'k-d') set(gca,'XTick',[0:NB]) set(gca,'XTickLabel','LNA';'Mixer';'Filter';'VGA';'ADC') grid on title('Sensitivity test') ylabel('Input SNR level in dB') legend('SNR',0) %%%% Adjacent channel test (signal is 6dB above sensitivity) %%%% Sig_level_IR = Signal_level_C2I; % desired signal level Int_level_IR = Interferer_level_C2I; % interference level at 3MHz offset for i=0:NB-1 Gain_accum(i+1) = sum(Gain_max(1:i)); % accumulated voltage gain from % antenna to block input end Int_gain=Gain_max - Atten_3M; % maximum gain for the interferer for i=0:NB-1 Int_gain_accum(i+1) = sum(Int_gain(1:i)); % accumulated voltage gain from end % antenna to block input Sig_level(1:NB) = Sig_level_IR+Gain_accum(1:NB); Int_level(1:NB) = Int_level_IR+Int_gain_accum(1:NB); figure plot([1:NB-1],Sig_level(2:end), 'k-o',[1:NB-1], Int_level(2:end), 'k-v') set(gca,'XTick',[1:NB-1]) set(gca,'XTickLabel','LNA';'Mixer';'Filter';'VGA';'ADC') grid on title('Adjacent channel test in Bluetooth mode') ylabel('Signal level in dBm') legend('Signal level','Interferer level',0)
222
%%% maximum signal test %%% Sig_level_IR = max_signal; for i=0:NB Gain_accum_min(i+1) = sum(Gain_min(1:i)); % accumulated voltage gain from end % antenna to block input Sig_level(1:NB) = Sig_level_IR+Gain_accum_min(1:NB); figure plot([0:NB-2],Sig_level(2:end), 'k-s') set(gca,'XTick',[0:NB-1]) set(gca,'XTickLabel','LNA';'Mixer';'Filter';'VGA';'ADC') grid on title('Maximum signal test in Bluetooth mode') ylabel('Signal level in dBm') legend('Signal level',0) %%% Signal at Sensitivity+51dB (LNA in Hi gain and VGA in Lo gain)%%% Sig_offset = 51; Sig_level_IR = Sensitivity+Sig_offset; Gain=[Gain_max(1:NB-2) Gain_min(NB-1:NB)]; NF=[NF_Gmax(1:NB-2) NF_Gmin(NB-1:NB)]; for i=0:NB Gain_accum(i+1) = sum(Gain(1:i)); % accumulated voltage gain from antenna End % to block input Sig_level(1:NB) = Sig_level_IR+Gain_accum(1:NB); NF_IR_Lin = (10.^(NF/10)-1)./10.^(Gain_accum(1:NB)/10)+1; % NF referred to the % Rx input in linear units Noise_level_IR = -174+10*log10(BW); % thermal noise at antenna for i=0:NB-1 NF_IR_accum_Lin(i+1)=sum(NF_IR_Lin(1:i)-1)+1; end Noise_level(1:NB) = 10*log10(10.^((Noise_level_IR+Gain_accum(1:NB))/10).*NF_IR_accum_Lin); figure plot([1:NB-1],Sig_level(2:end), 'k-o',[1:NB-1], Noise_level(2:end), 'k-s') set(gca,'XTick',[1:NB]) set(gca,'XTickLabel','LNA';'Mixer';'Filter';'VGA';'ADC') grid on title(' Gain switching test when LNA gain is Hi and VGA gain is Lo') ylabel('Input signal and noise levels in dBm') legend('Signal level','Noise level',0)
223
% Signal at Sensitivity+51dB (LNA in Lo gain and VGA in Lo gain) Sig_offset = 51; Sig_level_IR = Sensitivity+Sig_offset; Gain=Gain_min; NF=NF_Gmin; for i=0:NB Gain_accum(i+1) = sum(Gain(1:i)); % accumulated voltage gain from antenna End % to block input Sig_level(1:NB) = Sig_level_IR+Gain_accum(1:NB); NF_IR_Lin = (10.^(NF/10)-1)./10.^(Gain_accum(1:NB)/10)+1; % NF referred to the % Rx input in Noise_level_IR = -174+10*log10(BW); % thermal noise at antenna for i=0:NB-1 NF_IR_accum_Lin(i+1)=sum(NF_IR_Lin(1:i)-1)+1; end Noise_level(1:NB) = 10*log10(10.^((Noise_level_IR+Gain_accum(1:NB))/10).*NF_IR_accum_Lin); figure plot([1:NB-1],Sig_level(2:end), 'k-o',[1:NB-1], Noise_level(2:end), 'k-s') set(gca,'XTick',[1:NB]) set(gca,'XTickLabel','LNA';'Mixer';'Filter';'VGA';'ADC') grid on title(' Gain switching test when LNA gain is Lo and VGA gain is Lo') ylabel('Input signal and noise levels in dBm') legend('Signal level','Noise level',0)
224
D.2. Wi-Fi System Design MATLAB Code
%This file is for WiFi receiver system level design% %Author: Ahmed Emira% %Below is the flow chart of the system level design % _________________ __________________ % | Baseband | | Standard | % | Requirements | | Specifications | % | (SNRmin, C/Imin) | |_________________| % |_________________| / % \ / % \ / % \ / % \ / % _____________________ % | Derive required | % | system level specs | % |____________________| % | % | % __________\|/_________ % | Blocks’ specifications | _______ % |____________________| | % | % | | % __________\|/_________ | % | Calculate system specs | /|\ % |____________________| | % | | % ___________\|/___________ | NO % | Do we meet required specs |_____ | % |_______________________ | % | % | YES % _____\|/_____ % | Stop | % |___________ | % % All IIP3's are in dBm referenced to 50Ohm, unless otherwise specified % All NF's are in dB referenced to 50Ohm, unless otherwise specified % All gains are in dB unless otherwise specified
225
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%% Constants %%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% K = 1.38e-23; % Boltzmann constan T = 300; % room temperature %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%% Baseband Requirements %%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% Eb_No=13; % dB C2I=13; % minimum Carrier to inband interference ratio (dB) %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%% Standard specifications %%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% Sensitivity = -80; % dBm max_signal = -4; % dBm Rb = 11e6; %Mbps BW=6e6; %MHz % Adjacent channel specs Signal_level_C2I=-70; %dBm C2I_25MHz = -35; % dB Interferer_level_C2I = Signal_level_C2I-C2I_25MHz; %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%% General parameters %%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% Signal_ADC_max = 10; % this is the maximum signal level at ADC input(dBm) %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%% Required system specifications %%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% NF_margin = 6; Req_syst_NF = Sensitivity - (-174+10*log10(Rb)) - Eb_No - NF_margin IIP3_margin = 3; IM3 = Signal_level_C2I-C2I; Req_syst_IIP3 = (3*Interferer_level_C2I-IM3-37)/2+IIP3_margin % 37dB is subtracted since the interference % is a wideband CCK signal, not two tones. % No IP3 test is specified
226
Req_syst_IIP2 = 2*Interferer_level_C2I-Signal_level_C2I+C2I-20 % 20dB is subtracted since the interference % is a wideband CCK signal, not two tones. % No IP2 test is specified Req_syst_IIP2 = 2*Interferer_level_C2I-Signal_level_C2I+C2I-20 % 20dB is subtracted since the interference % is a wideband CCK signal, not two tones. Rx_Gain_max = Signal_ADC_max-Sensitivity; Rx_Gain_min = Signal_ADC_max-max_signal; %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%% Block specifications %%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %indices [RF_filter = 1, LNA = 2, Mixer = 3, Filter = 4, VGA = 5] NB=6; %number of blocks RF_filer = 1; %indeces LNA = 2; Mixer = 3; Filter = 4; VGA = 5; ADC = 6; ADC_SNR = 42; % dB, assumed for now! ADC_NF = Signal_ADC_max - ADC_SNR - 10*log10(4*K*T*BW/2*1000); % RF_filter LNA Mixer Filter VGA ADC IIP2 = [ 100 11 48 64 31 41] % in dBm IIP3 = [ 100 -8 5 23 10 10] % in dBm NF_Gmax = [ 0 3 15 32 30 ADC_NF] % at max gain,in dBNF_Gmin = [ 0 15 15 32 40 ADC_NF] % at min gain in dB Gain_max = [ 0 15 16 6 43 0 ] % max gain in dB Gain_min = [ 0 -15 16 6 0 0 ] % min gain in dB Atten_25M =[0 0 0 50 0 0 ] % interference % attenuation % relative to the % max gain ADC_ENOB = 6; % assumed for now! Mixer_feedthrough = -30; %RF to IF isolation in dB %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%% LO phase noise %%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% LO_PN = -Interferer_level_C2I-Eb_No-10*log10(Rb)+10*log10(10^(Signal_level_C2I/10)-10^(Sensitivity/10))-NF_margin % Phase noise in dBc/Hz at 25MHz
227
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%% System specifications %%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% for i=0:NB Gain_max_accum(i+1) = sum(Gain_max(1:i)); % accumulated voltage gain from antenna % to block input end for i=0:NB-1 Int_atten_accum(i+1) = sum(Atten_25M(1:i)); % accumulated gain from antenna end % to block input for the interferer IIP3_eff = IIP3+3*Int_atten_accum; % effective IP3 IIP2_eff = IIP2+2*Int_atten_accum; % effective IP2 % IM2 of the LNA is attenuated by the mixer's feedthrough IIP2_eff(LNA) = IIP2_eff(LNA) - (Mixer_feedthrough - Gain_max(Mixer)); % IIP2 is done when the signal is -70dBm so VGA gain is reduced by -70-(-80)=10dBGain_IIP2 = Gain_max_accum(1:NB)+[0 0 0 0 0 -10]; IIP2_IR_Lin = 10.^((IIP2_eff-Gain_IIP2)/20); % IIP2 referred to the Rx input in % sqrt(mW) IIP3_IR_Lin = 10.^((IIP3_eff-Gain_max_accum(1:NB))/10); % IIP3 referred to the % Rx input in sqrt(mW) NF_IR_Lin = (10.^(NF_Gmax/10)-1)./10.^(Gain_max_accum(1:NB)/10)+1; % NF referred to Rx input in % linear units Syst_NF = 10*log10(sum(NF_IR_Lin-1)+1) % System NF in dB Syst_IIP2 = 20*log10(1/sum(1./IIP2_IR_Lin)) % System IIP2 in dBm Syst_IIP3 = 10*log10(1/sum(1./IIP3_IR_Lin)) % System IIP3 in dBm %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%% Test scenarios %%%%%%%%%%%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%% Sensitivity test %%% Sig_level_IR = Sensitivity; % signal level at antenna (Rx input referred) Noise_level_IR = -174+10*log10(BW); % thermal noise at antenna Sig_level(1:NB) = Sig_level_IR+Gain_max_accum(1:NB); % signal level at each % block input
228
for i=0:NB-1 NF_IR_accum_Lin(i+1)=sum(NF_IR_Lin(1:i)-1)+1; end %Noise_level(1) = Noise_level_IR; Noise_level(1:NB) = 10*log10(10.^((Noise_level_IR+Gain_max_accum(1:NB))/10).*NF_IR_accum_Lin); SNR = Sig_level-Noise_level; figure plot([0:NB-2],Sig_level(2:end), 'k-o',[0:NB-2], Noise_level(2:end), 'k-s') set(gca,'XTick',[0:NB]) set(gca,'XTickLabel','LNA';'Mixer';'Filter';'VGA';'ADC') grid on title('Sensitivity test') ylabel('Input signal and noise levels in dBm') legend('Signal level','Noise level',0) figure plot([0:NB-2], SNR(2:end), 'k-d') set(gca,'XTick',[0:NB]) set(gca,'XTickLabel','LNA';'Mixer';'Filter';'VGA';'ADC') grid on title('Sensitivity test') ylabel('Input SNR level in dB') legend('SNR',0) %%% Adjacent channel test (signal is 6dB above sensitivity) %%% Sig_level_IR = Signal_level_C2I; % desired signal level Int_level_IR = Interferer_level_C2I; % interference level at 25MHz offset Gain=Gain_max; Gain(VGA)=Gain_max(VGA)-(Sig_level_IR-Sensitivity) for i=0:NB-1 Gain_accum(i+1) = sum(Gain(1:i)); % accumulated voltage gain from antenna end % to block input Int_gain=Gain - Atten_25M; % maximum gain for the interferer for i=0:NB-1 Int_gain_accum(i+1) = sum(Int_gain(1:i)); % accumulated voltage gain from end % antenna to block input Sig_level(1:NB) = Sig_level_IR+Gain_accum(1:NB); Int_level(1:NB) = Int_level_IR+Int_gain_accum(1:NB);
229
figure plot([1:NB-1],Sig_level(2:end), 'k-o',[1:NB-1], Int_level(2:end), 'k-v') set(gca,'XTick',[1:NB-1]) set(gca,'XTickLabel','LNA';'Mixer';'Filter';'VGA';'ADC') grid on title('Adjacent channel test in WiFi mode') ylabel('Signal level in dBm') legend('Signal level','Interferer level',0) %%% maximum signal test %%% Sig_level_IR = max_signal; for i=0:NB Gain_accum_min(i+1) = sum(Gain_min(1:i)); % accumulated voltage gain from end % antenna to block input Sig_level(1:NB) = Sig_level_IR+Gain_accum_min(1:NB); figure plot([0:NB-2],Sig_level(2:end), 'k-s') set(gca,'XTick',[0:NB-1]) set(gca,'XTickLabel','LNA';'Mixer';'Filter';'VGA';'ADC') grid on title('Maximum signal test in WiFi mode') ylabel('Signal level in dBm') legend('Signal level',0) %%% Signal at Sensitivity+20dB (the LNA in Hi gain) %%% Sig_offset = -37-Sensitivity; Sig_level_IR = Sensitivity+Sig_offset; Gain=Gain_max; Gain(VGA)=Gain_max(VGA)-Sig_offset; NF=NF_Gmax; NF(VGA)=NF_Gmin(VGA); for i=0:NB Gain_accum(i+1) = sum(Gain(1:i)); % accumulated voltage gain from antenna End % to block input Sig_level(1:NB) = Sig_level_IR+Gain_accum(1:NB); NF_IR_Lin = (10.^(NF/10)-1)./10.^(Gain_accum(1:NB)/10)+1; % NF referred to the % Rx input in linear units Noise_level_IR = -174+10*log10(BW); % thermal noise at antenna for i=0:NB-1 NF_IR_accum_Lin(i+1)=sum(NF_IR_Lin(1:i)-1)+1; end
230
Noise_level(1:NB) = 10*log10(10.^((Noise_level_IR+Gain_accum(1:NB))/10).*NF_IR_accum_Lin); figure plot([1:NB-1],Sig_level(2:end), 'k-o',[1:NB-1], Noise_level(2:end), 'k-s') set(gca,'XTick',[1:NB]) set(gca,'XTickLabel','LNA';'Mixer';'Filter';'VGA';'ADC') grid on title(' Gain switching test when LNA gain is Hi and VGA gain is Lo') ylabel('Input signal and noise levels in dBm') legend('Signal level','Noise level',0) %%% Signal at Sensitivity+20dB (the LNA in Lo gain) %%% Sig_offset = -37-Sensitivity; Sig_level_IR = Sensitivity+Sig_offset; Gain=Gain_min; Gain(VGA)=Gain_max(VGA)-Sig_offset+Gain_max(LNA)-Gain_min(LNA); NF=NF_Gmin; for i=0:NB Gain_accum(i+1) = sum(Gain(1:i)); % accumulated voltage gain from antenna end % to block input Sig_level(1:NB) = Sig_level_IR+Gain_accum(1:NB); NF_IR_Lin = (10.^(NF/10)-1)./10.^(Gain_accum(1:NB)/10)+1; % NF referred to the
% Rx input in linear % units Noise_level_IR = -174+10*log10(BW); % thermal noise at antenna for i=0:NB-1 NF_IR_accum_Lin(i+1)=sum(NF_IR_Lin(1:i)-1)+1; end Noise_level(1:NB) = 10*log10(10.^((Noise_level_IR+Gain_accum(1:NB))/10).*NF_IR_accum_Lin); figure plot([1:NB-1],Sig_level(2:end), 'k-o',[1:NB-1], Noise_level(2:end), 'k-s') set(gca,'XTick',[1:NB]) set(gca,'XTickLabel','LNA';'Mixer';'Filter';'VGA';'ADC') grid on title(' Gain switching test when LNA gain is Lo and VGA gain is Lo') ylabel('Input signal and noise levels in dBm') legend('Signal level','Noise level',0) %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% %%%%%% IIP2, IIP3, and Noise contributions of different blocks %%%%%%% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% IIP2_contributions = 10^(Syst_IIP2/20)./IIP2_IR_Lin(2:end);
231
figure LABELS_IIP2 = 'Mixer','Filter', 'LNA+VGA+ADC'; pie([IIP2_contributions(Mixer-1) IIP2_contributions(Filter-1) IIP2_contributions(LNA-1)+IIP2_contributions(VGA-1)+ IIP2_contributions(ADC-1)], LABELS_IIP2); title('IIP2 contributions of different blocks') IIP3_contributions = 10.^(Syst_IIP3/10)./IIP3_IR_Lin(2:end); figure LABELS_IIP3 = 'LNA','Mixer','Filter', 'VGA', 'ADC'; pie(IIP3_contributions, LABELS_IIP3); title('IIP3 contributions of different blocks') NF_contributions = (NF_IR_Lin-1)/(10.^(Syst_NF/10) -1); LABELS_NF = 'RF Filter','LNA','Mixer','Filter', 'VGA', 'SD ADC'; figure pie(NF_contributions, LABELS_NF); title('Noise contributions of different blocks') %%% Power consumption and area contributions in WiFi mode %%% Power_contributions_WiFi = [29.2 26.9 5.8 4.5 33.5]; figure LABELS_power = 'LNA/Mixer','PLL', 'Filter', 'VGA', 'ADC'; pie(Power_contributions_WiFi, LABELS_power); title('Power consumption contributions of different blocks in WiFi mode') Area_contributions_WiFi = [4.5 13.2 15.9 8.1 58.3]; figure LABELS_area = 'LNA/Mixer','PLL', 'Filter', 'VGA', 'ADC'; pie(Area_contributions_WiFi, LABELS_area); title('Area contributions of different blocks in WiFi mode')
232
APPENDIX E
LIST OF PAPERS USED IN THIS DISSERTATION
E.1. Journal Papers
1. Ahmed N. Mohielding, Ahmed Emira, Edgar Sánchez-Sinencio, “A 100MHz,
8mW ROM-Less Quadrature Direct Digital Frequency Synthesizer,” IEEE J.
Solid-State Circuits, vol. 37, no. 10, pp. 1235-1243, Oct. 2002.
2. Wenjun Sheng, Bo Xia, Ahmed Emira, Chunyu Xin, Ari Y. Valero-López, Sung
T. Moon, and Edgar Sánchez-Sinencio, “A 3V, 0.35µm CMOS Bluetooth
Receiver IC,” IEEE J. Solid-State Circuits, vol. 38, no. 1, pp. 30-42, Jan. 2003.
3. Ahmed Emira and Edgar Sánchez-Sinencio, “A Low-Power CMOS Complex
Filter with Frequency Tuning,” IEEE Trans. Circuits and Syst. II, vol. 50, no. 10,
pp. 742-754, Oct. 2003.
4. Ahmed Emira, Alberto Valdes-Garcia, Bo Xia, Ahmed Mohieldin, Ari Valero-
Lopez, Sung Moon, Chunyu Xin, and Edgar Sánchez-Sinencio “Chameleon: A
Dual-Mode 802.11b/Bluetooth Receiver in 0.25µm BiCMOS,” to be submitted to
IEEE J. of Solid-State Circuits.
5. Ahmed Emira and Edgar Sánchez-Sinencio, “Variable Gain Amplifier with Offset
Cancellation,” to be submitted to IEEE J. of Solid-State Circuits.
233
E.2. Conference Papers
1. Ahmed Emira, Alberto Valdes-Garcia, Bo Xia, Ahmed Mohieldin, Ari Valero-
Lopez, Sung Moon, Chunyu Xin, and Edgar Sánchez-Sinencio . “A Dual-Mode
802.11b/Bluetooth Receiver in 0.25µm BiCMOS,” in IEEE Int. Solid-State
Circuits Conf. (ISSCC) Dig. Tech. Papers, San Francisco, CA, Feb. 2004.
2. Ahmed Emira, Alberto Valdes-Garcia, and Edgar Sánchez-Sinencio, “System
Level WLAN Receiver Design,” accepted in MidWest Symp. on Circuits and
Syst., Cairo, Egypt, Dec. 2003.
3. Ahmed Emira and Edgar Sánchez-Sinencio, “A Low-Power Complex Filter for
Bluetooth with Frequency Tuning,” IEEE Int. Symp. Circuits and Systems, ISCAS
2003, Bangkok, Thailand, May 2003.
4. Ahmed Emira and Edgar Sánchez-Sinencio, “Variable Gain Amplifier with
Offset Cancellation,” Great Lakes Symp. on VLSI, GLSVLSI 2003, Washington
DC, April 2003.
5. Wenjun Sheng, Bo Xia, Ahmed Emira, and Edgar Sánchez-Sinencio, “Integrated
Receiver Design for Bluetooth”, IEEE CAS-Notre Dame Workshop on Wireless
Communications and Networking, University of Notre Dame, Aug. 2001.
6. Ahmed Emira and Edgar Sánchez-Sinencio, “Design Tradeoffs of CMOS
Current Mirrors Using One-Equation for All-Region Model,” IEEE Int. Symp.
Circuits and Systems, ISCAS 2002, Scottsdale, Arizona, May 2002.
234
7. Ahmed N. Mohieldin, Ahmed Emira, Edgar Sánchez-Sinencio, “A 2v 11 Bit
Incremental A/D Converter Using Floating Gate Technique,” IEEE Int. Symp.
Circuits and Systems, ISCAS 2002, Scottsdale, Arizona, May 2002.
8. Wenjun Sheng, Bo Xia, Ahmed Emira, Chunyu Xin, Ari Y. Valero-López, Sung
T. Moon, and Edgar Sánchez-Sinencio, “A Fully-Integrated 0.35µm CMOS
Bluetooth Receiver,” IEEE Custom Integrated Circuits Conference, CICC 2002,
Orlando, Florida, May 2002.
235
VITA
Ahmed A. Emira was born in Cairo, Egypt in 1974. He received the B.Eng. and
M.Eng. degrees from the Electronics and Communications Department, Cairo University,
Egypt, in 1997 and 1999, respectively. He worked as a teaching assistant in the same
department from July 1997 until December 1999.
Mr. Emira worked towards his Ph.D. degree at the Analog and Mixed Signal Center,
Department of Electrical Engineering, Texas A&M University, College Station, TX,
since January 2000 and received his Ph.D. in December 2003. He held an internship at
Motorola Inc., Austin, TX, from September 2001 till August 2002. His current research
interests include mixed-signal circuits, analog filter design, and communications systems
architectures.
Permanent address: 7001 South Congress Ave. #F202, Austin, TX 78745.